Port of r200 VB code. VB path should be as fast as the immediate path now, tuxracer should render correctly. Immediate path was left enabled.

8 files changed, 574 insertions, 156 deletions

diff --git a/src/mesa/drivers/dri/r300/Makefile b/src/mesa/drivers/dri/r300/Makefile
index f76ffd974d..1e7a08452c 100644
--- a/src/mesa/drivers/dri/r300/Makefile
+++ b/src/mesa/drivers/dri/r300/Makefile
@@ -36,6 +36,7 @@ DRIVER_SOURCES = \
 		 r300_tex.c \
 		 r300_texstate.c \
 		 r300_vertexprog.c \
+		 r300_maos.c \
 		 \
 		 r200_context.c \
 		 r200_ioctl.c \
diff --git a/src/mesa/drivers/dri/r300/r300_cmdbuf.c b/src/mesa/drivers/dri/r300/r300_cmdbuf.c
index 99c1f23dfd..0434421a49 100644
--- a/src/mesa/drivers/dri/r300/r300_cmdbuf.c
+++ b/src/mesa/drivers/dri/r300/r300_cmdbuf.c
@@ -645,36 +645,31 @@ void r300EmitWait(r300ContextPtr rmesa, GLuint flags)
 	}
 }
 
-void r300EmitLOAD_VBPNTR(r300ContextPtr rmesa, int start)
+void r300EmitAOS(r300ContextPtr rmesa, GLuint nr, GLuint offset)
 {
-int i, a, count;
-GLuint dw;
-LOCAL_VARS
-
-count=rmesa->state.aos_count;
-
-a=1+(count>>1)*3+(count & 1)*2;
-start_packet3(RADEON_CP_PACKET3_3D_LOAD_VBPNTR, a-1);
-e32(count);
-for(i=0;i+1<count;i+=2){
-	e32(  (rmesa->state.aos[i].element_size << 0) 
-	     |(rmesa->state.aos[i].stride << 8)
-	     |(rmesa->state.aos[i+1].element_size << 16)
-	     |(rmesa->state.aos[i+1].stride << 24)
-	    );
-	e32(rmesa->state.aos[i].offset+start*4*rmesa->state.aos[i].stride);
-	e32(rmesa->state.aos[i+1].offset+start*4*rmesa->state.aos[i+1].stride);
-	}
-if(count & 1){
-	e32(  (rmesa->state.aos[count-1].element_size << 0) 
-	     |(rmesa->state.aos[count-1].stride << 8)
-	    );
-	e32(rmesa->state.aos[count-1].offset+start*4*rmesa->state.aos[count-1].stride);	
-	}
+	if (RADEON_DEBUG & DEBUG_VERTS)
+	    fprintf(stderr, "%s: nr=%d, ofs=0x%08x\n", __func__, nr, offset);
+    int sz = 1 + (nr >> 1) * 3 + (nr & 1) * 2;
+    int i;
+    LOCAL_VARS
+
+    start_packet3(RADEON_CP_PACKET3_3D_LOAD_VBPNTR, sz-1);
+    e32(nr);
+    for(i=0;i+1<nr;i+=2){
+        e32(  (rmesa->state.aos[i].aos_size << 0)
+             |(rmesa->state.aos[i].aos_stride << 8)
+             |(rmesa->state.aos[i+1].aos_size << 16)
+             |(rmesa->state.aos[i+1].aos_stride << 24)
+        );
+        e32(rmesa->state.aos[i].aos_offset+offset*4*rmesa->state.aos[i].aos_stride);
+        e32(rmesa->state.aos[i+1].aos_offset+offset*4*rmesa->state.aos[i+1].aos_stride);
+    }
+    if(nr & 1){
+        e32(  (rmesa->state.aos[nr-1].aos_size << 0)
+             |(rmesa->state.aos[nr-1].aos_stride << 8)
+        );
+        e32(rmesa->state.aos[nr-1].aos_offset+offset*4*rmesa->state.aos[nr-1].aos_stride);
+    }
 
-/* delay ? */
-#if 0
-e32(RADEON_CP_PACKET2);
-e32(RADEON_CP_PACKET2);
-#endif
 }
+
diff --git a/src/mesa/drivers/dri/r300/r300_cmdbuf.h b/src/mesa/drivers/dri/r300/r300_cmdbuf.h
index 7f30f15562..b471f2ac46 100644
--- a/src/mesa/drivers/dri/r300/r300_cmdbuf.h
+++ b/src/mesa/drivers/dri/r300/r300_cmdbuf.h
@@ -46,6 +46,9 @@ extern void r300EmitState(r300ContextPtr r300);
 extern void r300InitCmdBuf(r300ContextPtr r300);
 extern void r300DestroyCmdBuf(r300ContextPtr r300);
 
+extern void r300EmitAOS(r300ContextPtr rmesa, GLuint nr, GLuint offset);
+
+
 /**
  * Make sure that enough space is available in the command buffer
  * by flushing if necessary.
diff --git a/src/mesa/drivers/dri/r300/r300_context.h b/src/mesa/drivers/dri/r300/r300_context.h
index bd15d76a5b..b6a80ae005 100644
--- a/src/mesa/drivers/dri/r300/r300_context.h
+++ b/src/mesa/drivers/dri/r300/r300_context.h
@@ -110,9 +110,20 @@ struct r300_dma_region {
 	struct r300_dma_buffer *buf;
 	char *address;		/* == buf->address */
 	int start, end, ptr;	/* offsets from start of buf */
-	int aos_start;
-	int aos_stride;
-	int aos_size;
+
+    int aos_offset;     /* address in GART memory */
+    int aos_stride;     /* distance between elements, in dwords */
+    int aos_size;       /* number of components (1-4) */
+    int aos_format;     /* format of components */
+    int aos_reg;        /* VAP register assignment */
+
+/* Left here so immediate mode still works unmodified */
+	int element_size;
+	int stride;
+	int offset;
+	int ncomponents;
+	int reg;
+	int format;
 };
 
 struct r300_dma {
@@ -653,8 +664,12 @@ struct r300_state {
 	struct r300_vap_reg_state vap_reg;
 	struct r300_vertex_shader_state vertex_shader;
 	struct r300_pixel_shader_state pixel_shader;
-	struct r300_aos_rec aos[R300_MAX_AOS_ARRAYS];
+
+	struct r300_dma_region aos[R300_MAX_AOS_ARRAYS];
 	int aos_count;
+
+	GLuint *Elts;
+
 	GLuint render_inputs; /* actual render inputs that R300 was configured for. 
 				 They are the same as tnl->render_inputs for fixed pipeline */
 
diff --git a/src/mesa/drivers/dri/r300/r300_ioctl.c b/src/mesa/drivers/dri/r300/r300_ioctl.c
index 6ba5bf257a..8b022067f2 100644
--- a/src/mesa/drivers/dri/r300/r300_ioctl.c
+++ b/src/mesa/drivers/dri/r300/r300_ioctl.c
@@ -453,13 +453,13 @@ void r300ReleaseDmaRegion(r300ContextPtr rmesa,
 		if (RADEON_DEBUG & (DEBUG_IOCTL | DEBUG_DMA))
 			fprintf(stderr, "%s -- DISCARD BUF %d\n", __FUNCTION__,
 				region->buf->buf->idx);
-
 		cmd =
 		    (drm_radeon_cmd_header_t *) r300AllocCmdBuf(rmesa,
-								sizeof(*cmd),
+								sizeof(*cmd) / 4,
 								__FUNCTION__);
-		cmd->dma.cmd_type = RADEON_CMD_DMA_DISCARD;
+		cmd->dma.cmd_type = R300_CMD_DMA_DISCARD;
 		cmd->dma.buf_idx = region->buf->buf->idx;
+		
 		FREE(region->buf);
 		rmesa->dma.nr_released_bufs++;
 	}
diff --git a/src/mesa/drivers/dri/r300/r300_maos.c b/src/mesa/drivers/dri/r300/r300_maos.c
new file mode 100644
index 0000000000..a6c15e5220
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/r300_maos.c
@@ -0,0 +1,444 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/r300/r300_maos_arrays.c,v 1.3 2003/02/23 23:59:01 dawes Exp $ */
+/*
+Copyright (C) The Weather Channel, Inc.  2002.  All Rights Reserved.
+
+The Weather Channel (TM) funded Tungsten Graphics to develop the
+initial release of the Radeon 8500 driver under the XFree86 license.
+This notice must be preserved.
+
+Permission is hereby granted, free of charge, to any person obtaining
+a copy of this software and associated documentation files (the
+"Software"), to deal in the Software without restriction, including
+without limitation the rights to use, copy, modify, merge, publish,
+distribute, sublicense, and/or sell copies of the Software, and to
+permit persons to whom the Software is furnished to do so, subject to
+the following conditions:
+
+The above copyright notice and this permission notice (including the
+next paragraph) shall be included in all copies or substantial
+portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#include "glheader.h"
+#include "mtypes.h"
+#include "colormac.h"
+#include "imports.h"
+#include "macros.h"
+
+#include "swrast_setup/swrast_setup.h"
+#include "math/m_translate.h"
+#include "tnl/tnl.h"
+#include "tnl/t_context.h"
+
+#include "r300_context.h"
+#include "radeon_ioctl.h"
+#include "r300_state.h"
+#include "r300_maos.h"
+
+#define DEBUG_ALL DEBUG_VERTS
+
+#if defined(USE_X86_ASM)
+#define COPY_DWORDS( dst, src, nr )					\
+do {									\
+	int __tmp;							\
+	__asm__ __volatile__( "rep ; movsl"				\
+			      : "=%c" (__tmp), "=D" (dst), "=S" (__tmp)	\
+			      : "0" (nr),				\
+			        "D" ((long)dst),			\
+			        "S" ((long)src) );			\
+} while (0)
+#else
+#define COPY_DWORDS( dst, src, nr )		\
+do {						\
+   int j;					\
+   for ( j = 0 ; j < nr ; j++ )			\
+      dst[j] = ((int *)src)[j];			\
+   dst += nr;					\
+} while (0)
+#endif
+
+static void emit_vec4(GLcontext * ctx,
+		      struct r300_dma_region *rvb,
+		      char *data, int stride, int count)
+{
+	int i;
+	int *out = (int *)(rvb->address + rvb->start);
+
+	if (RADEON_DEBUG & DEBUG_VERTS)
+		fprintf(stderr, "%s count %d stride %d\n",
+			__FUNCTION__, count, stride);
+
+	if (stride == 4)
+		COPY_DWORDS(out, data, count);
+	else
+		for (i = 0; i < count; i++) {
+			out[0] = *(int *)data;
+			out++;
+			data += stride;
+		}
+}
+
+static void emit_vec8(GLcontext * ctx,
+		      struct r300_dma_region *rvb,
+		      char *data, int stride, int count)
+{
+	int i;
+	int *out = (int *)(rvb->address + rvb->start);
+
+	if (RADEON_DEBUG & DEBUG_VERTS)
+		fprintf(stderr, "%s count %d stride %d\n",
+			__FUNCTION__, count, stride);
+
+	if (stride == 8)
+		COPY_DWORDS(out, data, count * 2);
+	else
+		for (i = 0; i < count; i++) {
+			out[0] = *(int *)data;
+			out[1] = *(int *)(data + 4);
+			out += 2;
+			data += stride;
+		}
+}
+
+static void emit_vec12(GLcontext * ctx,
+		       struct r300_dma_region *rvb,
+		       char *data, int stride, int count)
+{
+	int i;
+	int *out = (int *)(rvb->address + rvb->start);
+
+	if (RADEON_DEBUG & DEBUG_VERTS)
+		fprintf(stderr, "%s count %d stride %d out %p data %p\n",
+			__FUNCTION__, count, stride, (void *)out, (void *)data);
+
+	if (stride == 12)
+		COPY_DWORDS(out, data, count * 3);
+	else
+		for (i = 0; i < count; i++) {
+			out[0] = *(int *)data;
+			out[1] = *(int *)(data + 4);
+			out[2] = *(int *)(data + 8);
+			out += 3;
+			data += stride;
+		}
+}
+
+static void emit_vec16(GLcontext * ctx,
+		       struct r300_dma_region *rvb,
+		       char *data, int stride, int count)
+{
+	int i;
+	int *out = (int *)(rvb->address + rvb->start);
+
+	if (RADEON_DEBUG & DEBUG_VERTS)
+		fprintf(stderr, "%s count %d stride %d\n",
+			__FUNCTION__, count, stride);
+
+	if (stride == 16)
+		COPY_DWORDS(out, data, count * 4);
+	else
+		for (i = 0; i < count; i++) {
+			out[0] = *(int *)data;
+			out[1] = *(int *)(data + 4);
+			out[2] = *(int *)(data + 8);
+			out[3] = *(int *)(data + 12);
+			out += 4;
+			data += stride;
+		}
+}
+
+static void emit_vector(GLcontext * ctx,
+			struct r300_dma_region *rvb,
+			char *data, int size, int stride, int count)
+{
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+
+	if (RADEON_DEBUG & DEBUG_VERTS)
+		fprintf(stderr, "%s count %d size %d stride %d\n",
+			__FUNCTION__, count, size, stride);
+
+	assert(!rvb->buf);
+
+	if (stride == 0) {
+		r300AllocDmaRegion(rmesa, rvb, size * 4, 4);
+		count = 1;
+		rvb->aos_offset	= GET_START(rvb);
+		rvb->aos_stride	= 0;
+		rvb->aos_size	= size;
+	} else {
+		r300AllocDmaRegion(rmesa, rvb, size * count * 4, 4);	/* alignment? */
+		rvb->aos_offset	= GET_START(rvb);
+		rvb->aos_stride	= size;
+		rvb->aos_size	= size;
+	}
+
+	/* Emit the data
+	 */
+	switch (size) {
+	case 1:
+		emit_vec4(ctx, rvb, data, stride, count);
+		break;
+	case 2:
+		emit_vec8(ctx, rvb, data, stride, count);
+		break;
+	case 3:
+		emit_vec12(ctx, rvb, data, stride, count);
+		break;
+	case 4:
+		emit_vec16(ctx, rvb, data, stride, count);
+		break;
+	default:
+		assert(0);
+		exit(1);
+		break;
+	}
+
+}
+
+/* Emit any changed arrays to new GART memory, re-emit a packet to
+ * update the arrays.
+ */
+void r300EmitArrays(GLcontext * ctx, GLuint inputs)
+{
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+	r300ContextPtr r300 = rmesa;
+	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;
+
+/*FIXME: handle vertex program input */
+	if (inputs & _TNL_BIT_POS) {
+		if (RADEON_DEBUG & DEBUG_ALL)
+			fprintf(stderr, "[%d] _TNL_BIT_POS: sz=%d, st=%d, c=%d\n",
+				nr, VB->ObjPtr->size, VB->ObjPtr->stride, count);
+		emit_vector(ctx, &rmesa->state.aos[nr++],
+					(char *)VB->ObjPtr->data,
+					VB->ObjPtr->size,
+					VB->ObjPtr->stride, count);
+
+		vic_1 |= R300_INPUT_CNTL_POS;
+		rmesa->state.aos[nr-1].aos_reg	  = aa_vap_reg++;
+		rmesa->state.aos[nr-1].aos_format = AOS_FORMAT_FLOAT;
+	}
+
+	if (inputs & _TNL_BIT_NORMAL) {
+		if (RADEON_DEBUG & DEBUG_ALL)
+			fprintf(stderr, "[%d] _TNL_BIT_NORMAL: sz=%d, st=%d, c=%d\n",
+				nr, VB->NormalPtr->size, VB->NormalPtr->stride, count);
+		emit_vector(ctx, &rmesa->state.aos[nr++],
+				(char *)VB->NormalPtr->data,
+				VB->NormalPtr->size,
+				VB->NormalPtr->stride, count);
+
+		vic_1 |= R300_INPUT_CNTL_NORMAL;
+		rmesa->state.aos[nr-1].aos_reg	  = aa_vap_reg++;
+		rmesa->state.aos[nr-1].aos_format = AOS_FORMAT_FLOAT;
+	}
+
+	if (inputs & VERT_BIT_COLOR0) {
+		int emitsize;
+		if (RADEON_DEBUG & DEBUG_ALL)
+			fprintf(stderr, "[%d] _TNL_BIT_COLOR0: sz=%d, st=%d, c=%d\n",
+				nr, VB->ColorPtr[0]->size, VB->ColorPtr[0]->stride, count);
+
+		if (VB->ColorPtr[0]->size == 4 &&
+		    (VB->ColorPtr[0]->stride != 0 ||
+		     VB->ColorPtr[0]->data[0][3] != 1.0)) {
+			emitsize = 4;
+		} else {
+			emitsize = 3;
+		}
+
+		emit_vector(ctx,
+			    &(rmesa->state.aos[nr++]),
+			    (char *)VB->ColorPtr[0]->data,
+			    emitsize, VB->ColorPtr[0]->stride, count);
+//			    emitsize, VB->ColorPtr[0]->stride, count);
+
+		vic_1 |= R300_INPUT_CNTL_COLOR;
+		rmesa->state.aos[nr-1].aos_reg	  = aa_vap_reg++;
+		rmesa->state.aos[nr-1].aos_format = AOS_FORMAT_FLOAT_COLOR;
+	}
+
+	if (inputs & VERT_BIT_COLOR1) {
+		if (RADEON_DEBUG & DEBUG_ALL)
+			fprintf(stderr, "[%d] _TNL_BIT_COLOR1: sz=%d, st=%d, c=%d\n",
+				nr, VB->SecondaryColorPtr[0]->size, VB->SecondaryColorPtr[0]->stride, count);
+
+		emit_vector(ctx,
+			    &rmesa->state.aos[nr++],
+			    (char *)VB->SecondaryColorPtr[0]->data,
+			    VB->SecondaryColorPtr[0]->size, VB->SecondaryColorPtr[0]->stride, count);
+		rmesa->state.aos[nr-1].aos_reg	  = aa_vap_reg++;
+		rmesa->state.aos[nr-1].aos_format = AOS_FORMAT_FLOAT_COLOR;
+	}
+
+	for (i = 0; i < ctx->Const.MaxTextureUnits; i++) {
+		if (inputs & (_TNL_BIT_TEX0 << i)) {
+			if (RADEON_DEBUG & DEBUG_ALL)
+				fprintf(stderr, "[%d] _TNL_BIT_TEX%02d: sz=%d, st=%d, c=%d\n",
+					nr, i, VB->TexCoordPtr[i]->size, VB->TexCoordPtr[i]->stride, count);
+			emit_vector(ctx,
+				    &(rmesa->state.aos[nr++]),
+				    (char *)VB->TexCoordPtr[i]->data,
+				    VB->TexCoordPtr[i]->size,
+				    VB->TexCoordPtr[i]->stride, count);
+
+			vic_1 |= R300_INPUT_CNTL_TC0 << i;
+			rmesa->state.aos[nr-1].aos_reg	  = aa_vap_reg++;
+			rmesa->state.aos[nr-1].aos_format = AOS_FORMAT_FLOAT;
+		}
+	}
+
+int cmd_reserved=0;
+int cmd_written=0;
+drm_radeon_cmd_header_t *cmd = NULL;
+
+#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)->unchecked_state.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+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;
+
+		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)->unchecked_state.count = (nr+1)>>1;
+
+	/* Set up input_cntl */
+
+	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;
+
+#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]=R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT
+					| R300_VAP_OUTPUT_VTX_FMT_0__COLOR_PRESENT;
+
+	r300->hw.vof.cmd[R300_VOF_CNTL_1]=0;
+	for(i=0;i < ctx->Const.MaxTextureUnits;i++)
+		if(r300->state.render_inputs & (_TNL_BIT_TEX0<<i))
+			r300->hw.vof.cmd[R300_VOF_CNTL_1]|=(4<<(3*i));
+
+	rmesa->state.aos_count = nr;
+}
+
+void r300ReleaseArrays(GLcontext * ctx)
+{
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+	int i;
+
+	for (i=0;i<rmesa->state.aos_count;i++) {
+		r300ReleaseDmaRegion(rmesa, &rmesa->state.aos[i], __FUNCTION__);
+	}
+}
diff --git a/src/mesa/drivers/dri/r300/r300_maos.h b/src/mesa/drivers/dri/r300/r300_maos.h
new file mode 100644
index 0000000000..6f6cce2f46
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/r300_maos.h
@@ -0,0 +1,48 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/r200/r200_maos.h,v 1.1 2002/10/30 12:51:52 alanh Exp $ */
+/*
+Copyright (C) The Weather Channel, Inc.  2002.  All Rights Reserved.
+
+The Weather Channel (TM) funded Tungsten Graphics to develop the
+initial release of the Radeon 8500 driver under the XFree86 license.
+This notice must be preserved.
+
+Permission is hereby granted, free of charge, to any person obtaining
+a copy of this software and associated documentation files (the
+"Software"), to deal in the Software without restriction, including
+without limitation the rights to use, copy, modify, merge, publish,
+distribute, sublicense, and/or sell copies of the Software, and to
+permit persons to whom the Software is furnished to do so, subject to
+the following conditions:
+
+The above copyright notice and this permission notice (including the
+next paragraph) shall be included in all copies or substantial
+portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#ifndef __R300_MAOS_H__
+#define __R300_MAOS_H__
+
+#ifdef GLX_DIRECT_RENDERING
+
+#include "r300_context.h"
+
+extern void r300EmitArrays(GLcontext * ctx, GLuint inputs);
+extern void r300ReleaseArrays(GLcontext * ctx);
+
+#endif
+#endif
+
diff --git a/src/mesa/drivers/dri/r300/r300_render.c b/src/mesa/drivers/dri/r300/r300_render.c
index 56c8a5af4d..1d21574496 100644
--- a/src/mesa/drivers/dri/r300/r300_render.c
+++ b/src/mesa/drivers/dri/r300/r300_render.c
@@ -485,8 +485,8 @@ static GLboolean r300_run_immediate_render(GLcontext *ctx,
    efloat(1.0);
    #endif
    
-   /* We need LOAD_VBPNTR to setup AOS_ATTR fields.. the offsets are irrelevant */
-   r300EmitLOAD_VBPNTR(rmesa, 0);
+/* Why do we need this for immediate mode?? */
+//   r300EmitLOAD_VBPNTR(rmesa, 0);
    
    for(i=0; i < VB->PrimitiveCount; i++){
        GLuint prim = VB->Primitive[i].mode;
@@ -508,91 +508,8 @@ static GLboolean r300_run_immediate_render(GLcontext *ctx,
    return GL_FALSE;
 }
 
-
 /* vertex buffer implementation */
 
-/* We use the start part of GART texture buffer for vertices */
-
-
-static void upload_vertex_buffer(r300ContextPtr rmesa, GLcontext *ctx)
-{
-	TNLcontext *tnl = TNL_CONTEXT(ctx);
-	struct vertex_buffer *VB = &tnl->vb;
-	int idx=0;
-	int i,j,k;
-	radeonScreenPtr rsp=rmesa->radeon.radeonScreen;
-	GLuint render_inputs;
-	
-	/* A hack - we don't want to overwrite vertex buffers, so we
-	just use AGP space for them.. Fix me ! */
-	static int offset=0;
-	if(offset>2*1024*1024){
-		//fprintf(stderr, "Wrapping agp vertex buffer offset\n");
-		offset=0;
-		}
-	/* Not the most efficient implementation, but, for now, I just want something that
-	works */
-	/* to do - make single memcpy per column (is it possible ?) */
-	/* to do - use dirty flags to avoid redundant copies */
-	#define UPLOAD_VECTOR(v)\
-		{ \
-		/* Is the data dirty ? */ \
-		if (v->flags & ((1<<v->size)-1)) { \
-			/* fprintf(stderr, "size=%d vs stride=%d\n", v->size, v->stride); */ \
-			if(v->size*4==v->stride){\
-				/* fast path */  \
-				memcpy(rsp->gartTextures.map+offset, v->data, v->stride*VB->Count); \
-				} else { \
-				for(i=0;i<VB->Count;i++){ \
-					/* copy one vertex at a time*/ \
-					memcpy(rsp->gartTextures.map+offset+i*v->size*4, VEC_ELT(v, GLfloat, i), v->size*4); \
-					} \
-				} \
-			/* v->flags &= ~((1<<v->size)-1);*/ \
-			} \
-		rmesa->state.aos[idx].offset=rsp->gartTextures.handle+offset; \
-		offset+=v->size*4*VB->Count; \
-		idx++; \
-		}
-		
-   render_inputs = rmesa->state.render_inputs;
-
-   if(!render_inputs){
-   	WARN_ONCE("Aeiee ! render_inputs==0. Skipping rendering.\n");
-	return;
-   	}
-	/* coordinates */
-	if(render_inputs & _TNL_BIT_POS)
-		UPLOAD_VECTOR(VB->ObjPtr);
-	if(render_inputs & _TNL_BIT_NORMAL)
-		UPLOAD_VECTOR(VB->NormalPtr);
-	
-	/* color components */
-	if(render_inputs & _TNL_BIT_COLOR0)
-		UPLOAD_VECTOR(VB->ColorPtr[0]);
-	if(render_inputs & _TNL_BIT_COLOR1)
-		UPLOAD_VECTOR(VB->SecondaryColorPtr[0]);
-	
-	if(render_inputs & _TNL_BIT_FOG)
-		UPLOAD_VECTOR(VB->FogCoordPtr);
-	
-	/* texture coordinates */
-	for(k=0;k < ctx->Const.MaxTextureUnits;k++)
-		if(render_inputs & (_TNL_BIT_TEX0<<k))
-			UPLOAD_VECTOR(VB->TexCoordPtr[k]);
-	
-	if(render_inputs & _TNL_BIT_INDEX)
-		UPLOAD_VECTOR(VB->IndexPtr[0]);
-	if(render_inputs & _TNL_BIT_POINTSIZE)
-		UPLOAD_VECTOR(VB->PointSizePtr);
-
-	if(idx>=R300_MAX_AOS_ARRAYS){
-		fprintf(stderr, "Aieee ! Maximum AOS arrays count exceeded.. \n");
-		exit(-1);
-		}
-	//dump_inputs(ctx, render_inputs); return ;
-}
-
 static void r300_render_vb_primitive(r300ContextPtr rmesa, 
 	GLcontext *ctx,
 	int start,
@@ -624,50 +541,45 @@ static GLboolean r300_run_vb_render(GLcontext *ctx,
 		fprintf(stderr, "%s\n", __FUNCTION__);
 
    
-   reg_start(R300_RB3D_DSTCACHE_CTLSTAT,0);
+	r300ReleaseArrays(ctx);
+	r300EmitArrays(ctx, rmesa->state.render_inputs);
+
+//	LOCK_HARDWARE(&(rmesa->radeon));
+
+	reg_start(R300_RB3D_DSTCACHE_CTLSTAT,0);
 	e32(0x0000000a);
    
-   reg_start(0x4f18,0);
+	reg_start(0x4f18,0);
 	e32(0x00000003);
-   
-   r300_setup_routing(ctx, GL_FALSE);
-	
-   r300EmitState(rmesa);
 
-   /* setup array of structures data */
-   LOCK_HARDWARE(&(rmesa->radeon));
+	r300EmitState(rmesa);		
 
-   upload_vertex_buffer(rmesa, ctx);
-   //fprintf(stderr, "Using %d AOS arrays\n", n_arrays);
-   
-   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;
-       		
-	   /* We need LOAD_VBPNTR to setup AOS_ATTR fields.. */
-        r300EmitLOAD_VBPNTR(rmesa, start);
-       
-	r300_render_vb_primitive(rmesa, ctx, start, start + length, prim);
-   	}
-	
-    /* This sequence is required after any 3d drawing packet
-      I suspect it works around a bug (or deficiency) in hardware */
-  
-  reg_start(R300_RB3D_DSTCACHE_CTLSTAT,0);
+	rmesa->state.Elts = VB->Elts;
+
+	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;
+
+		r300EmitAOS(rmesa, rmesa->state.aos_count, start);   
+
+		r300_render_vb_primitive(rmesa, ctx, start, start + length, prim);
+	}
+
+	reg_start(R300_RB3D_DSTCACHE_CTLSTAT,0);
 	e32(0x0000000a);
-   
-   reg_start(0x4f18,0);
+  
+	reg_start(0x4f18,0);
 	e32(0x00000003);
-   
-   end_3d(PASS_PREFIX_VOID);
+
+//	end_3d(PASS_PREFIX_VOID);
    
    /* Flush state - we are done drawing.. */
-   r300FlushCmdBufLocked(rmesa, __FUNCTION__);
-   radeonWaitForIdleLocked(&(rmesa->radeon));
+//	r300FlushCmdBufLocked(rmesa, __FUNCTION__);
+//	radeonWaitForIdleLocked(&(rmesa->radeon));
    
-   UNLOCK_HARDWARE(&(rmesa->radeon));
-   return GL_FALSE;
+//	UNLOCK_HARDWARE(&(rmesa->radeon));
+	return GL_FALSE;
 }
 
 /**