mesa: merge gallium-0.2 into gallium-master-merge

Merge commit 'origin/gallium-0.2' into gallium-master-merge

Conflicts:

	Makefile
	docs/relnotes-7.4.html
	docs/relnotes.html
	src/mesa/drivers/dri/i965/brw_wm.h
	src/mesa/main/imports.c
	src/mesa/main/mtypes.h
	src/mesa/main/texcompress.c
	src/mesa/main/texenvprogram.c
	src/mesa/main/version.h
	src/mesa/vbo/vbo_exec_api.c
	src/mesa/vbo/vbo_save_draw.c

16 files changed, 3718 insertions, 0 deletions

diff --git a/src/gallium/drivers/nv20/Makefile b/src/gallium/drivers/nv20/Makefile
new file mode 100644
index 00000000000..d777fd3d8b4
--- /dev/null
+++ b/src/gallium/drivers/nv20/Makefile
@@ -0,0 +1,29 @@
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = nv20
+
+DRIVER_SOURCES = \
+	nv20_clear.c \
+	nv20_context.c \
+	nv20_fragprog.c \
+	nv20_fragtex.c \
+	nv20_miptree.c \
+	nv20_prim_vbuf.c \
+	nv20_screen.c \
+	nv20_state.c \
+	nv20_state_emit.c \
+	nv20_surface.c \
+	nv20_vbo.c
+#	nv20_vertprog.c
+
+C_SOURCES = \
+	$(COMMON_SOURCES) \
+	$(DRIVER_SOURCES)
+
+ASM_SOURCES = 
+
+include ../../Makefile.template
+
+symlinks:
+
diff --git a/src/gallium/drivers/nv20/nv20_clear.c b/src/gallium/drivers/nv20/nv20_clear.c
new file mode 100644
index 00000000000..29f4afd87c9
--- /dev/null
+++ b/src/gallium/drivers/nv20/nv20_clear.c
@@ -0,0 +1,13 @@
+#include "pipe/p_context.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_state.h"
+
+#include "nv20_context.h"
+
+void
+nv20_clear(struct pipe_context *pipe, struct pipe_surface *ps,
+	   unsigned clearValue)
+{
+	pipe->surface_fill(pipe, ps, 0, 0, ps->width, ps->height, clearValue);
+	ps->status = PIPE_SURFACE_STATUS_CLEAR;
+}
diff --git a/src/gallium/drivers/nv20/nv20_context.c b/src/gallium/drivers/nv20/nv20_context.c
new file mode 100644
index 00000000000..1659aec8fab
--- /dev/null
+++ b/src/gallium/drivers/nv20/nv20_context.c
@@ -0,0 +1,419 @@
+#include "draw/draw_context.h"
+#include "pipe/p_defines.h"
+#include "pipe/internal/p_winsys_screen.h"
+
+#include "nv20_context.h"
+#include "nv20_screen.h"
+
+static void
+nv20_flush(struct pipe_context *pipe, unsigned flags,
+	   struct pipe_fence_handle **fence)
+{
+	struct nv20_context *nv20 = nv20_context(pipe);
+
+	draw_flush(nv20->draw);
+
+	FIRE_RING(fence);
+}
+
+static void
+nv20_destroy(struct pipe_context *pipe)
+{
+	struct nv20_context *nv20 = nv20_context(pipe);
+
+	if (nv20->draw)
+		draw_destroy(nv20->draw);
+
+	FREE(nv20);
+}
+
+static void nv20_init_hwctx(struct nv20_context *nv20)
+{
+	struct nv20_screen *screen = nv20->screen;
+	struct nouveau_winsys *nvws = screen->nvws;
+	int i;
+	float projectionmatrix[16];
+	const boolean is_nv25tcl = (nv20->screen->kelvin->grclass == NV25TCL);
+
+	BEGIN_RING(kelvin, NV20TCL_DMA_NOTIFY, 1);
+	OUT_RING  (screen->sync->handle);
+	BEGIN_RING(kelvin, NV20TCL_DMA_TEXTURE0, 2);
+	OUT_RING  (nvws->channel->vram->handle);
+	OUT_RING  (nvws->channel->gart->handle); /* TEXTURE1 */
+	BEGIN_RING(kelvin, NV20TCL_DMA_COLOR, 2);
+	OUT_RING  (nvws->channel->vram->handle);
+	OUT_RING  (nvws->channel->vram->handle); /* ZETA */
+
+	BEGIN_RING(kelvin, NV20TCL_DMA_QUERY, 1);
+	OUT_RING  (0); /* renouveau: beef0351, unique */
+
+	BEGIN_RING(kelvin, NV20TCL_RT_HORIZ, 2);
+	OUT_RING  (0);
+	OUT_RING  (0);
+
+	BEGIN_RING(kelvin, NV20TCL_VIEWPORT_CLIP_HORIZ(0), 1);
+	OUT_RING  ((0xfff << 16) | 0x0);
+	BEGIN_RING(kelvin, NV20TCL_VIEWPORT_CLIP_VERT(0), 1);
+	OUT_RING  ((0xfff << 16) | 0x0);
+
+	for (i = 1; i < NV20TCL_VIEWPORT_CLIP_HORIZ__SIZE; i++) {
+		BEGIN_RING(kelvin, NV20TCL_VIEWPORT_CLIP_HORIZ(i), 1);
+		OUT_RING  (0);
+		BEGIN_RING(kelvin, NV20TCL_VIEWPORT_CLIP_VERT(i), 1);
+		OUT_RING  (0);
+	}
+
+	BEGIN_RING(kelvin, NV20TCL_VIEWPORT_CLIP_MODE, 1);
+	OUT_RING  (0);
+
+	BEGIN_RING(kelvin, 0x17e0, 3);
+	OUT_RINGf (0.0);
+	OUT_RINGf (0.0);
+	OUT_RINGf (1.0);
+
+	if (is_nv25tcl) {
+		BEGIN_RING(kelvin, NV20TCL_TX_RCOMP, 1);
+		OUT_RING  (NV20TCL_TX_RCOMP_LEQUAL | 0xdb0);
+	} else {
+		BEGIN_RING(kelvin, 0x1e68, 1);
+		OUT_RING  (0x4b800000); /* 16777216.000000 */
+		BEGIN_RING(kelvin, NV20TCL_TX_RCOMP, 1);
+		OUT_RING  (NV20TCL_TX_RCOMP_LEQUAL);
+	}
+
+	BEGIN_RING(kelvin, 0x290, 1);
+	OUT_RING  ((0x10 << 16) | 1);
+	BEGIN_RING(kelvin, 0x9fc, 1);
+	OUT_RING  (0);
+	BEGIN_RING(kelvin, 0x1d80, 1);
+	OUT_RING  (1);
+	BEGIN_RING(kelvin, 0x9f8, 1);
+	OUT_RING  (4);
+	BEGIN_RING(kelvin, 0x17ec, 3);
+	OUT_RINGf (0.0);
+	OUT_RINGf (1.0);
+	OUT_RINGf (0.0);
+
+	if (is_nv25tcl) {
+		BEGIN_RING(kelvin, 0x1d88, 1);
+		OUT_RING  (3);
+
+		BEGIN_RING(kelvin, NV25TCL_DMA_IN_MEMORY9, 1);
+		OUT_RING  (nvws->channel->vram->handle);
+		BEGIN_RING(kelvin, NV25TCL_DMA_IN_MEMORY8, 1);
+		OUT_RING  (nvws->channel->vram->handle);
+	}
+	BEGIN_RING(kelvin, NV20TCL_DMA_FENCE, 1);
+	OUT_RING  (0);	/* renouveau: beef1e10 */
+
+	BEGIN_RING(kelvin, 0x1e98, 1);
+	OUT_RING  (0);
+#if 0
+	if (is_nv25tcl) {
+		BEGIN_RING(NvSub3D, NV25TCL_DMA_IN_MEMORY4, 2);
+		OUT_RING  (NvDmaTT);	/* renouveau: beef0202 */
+		OUT_RING  (NvDmaFB);	/* renouveau: beef0201 */
+
+		BEGIN_RING(NvSub3D, NV20TCL_DMA_TEXTURE1, 1);
+		OUT_RING  (NvDmaTT);	/* renouveau: beef0202 */
+	}
+#endif
+	BEGIN_RING(kelvin, NV20TCL_NOTIFY, 1);
+	OUT_RING  (0);
+
+	BEGIN_RING(kelvin, 0x120, 3);
+	OUT_RING  (0);
+	OUT_RING  (1);
+	OUT_RING  (2);
+
+/* error: ILLEGAL_MTHD, PROTECTION_FAULT
+	BEGIN_RING(kelvin, NV20TCL_VIEWPORT_TRANSLATE_X, 4);
+	OUT_RINGf (0.0);
+	OUT_RINGf (512.0);
+	OUT_RINGf (0.0);
+	OUT_RINGf (0.0);
+*/
+
+	if (is_nv25tcl) {
+		BEGIN_RING(kelvin, 0x022c, 2);
+		OUT_RING  (0x280);
+		OUT_RING  (0x07d28000);
+	}
+
+/* * illegal method, protection fault
+	BEGIN_RING(NvSub3D, 0x1c2c, 1);
+	OUT_RING  (0); */
+
+	if (is_nv25tcl) {
+		BEGIN_RING(kelvin, 0x1da4, 1);
+		OUT_RING  (0);
+	}
+
+/* * crashes with illegal method, protection fault
+	BEGIN_RING(NvSub3D, 0x1c18, 1);
+	OUT_RING  (0x200); */
+
+	BEGIN_RING(kelvin, NV20TCL_RT_HORIZ, 2);
+	OUT_RING  ((0 << 16) | 0);
+	OUT_RING  ((0 << 16) | 0);
+
+	/* *** Set state *** */
+
+	BEGIN_RING(kelvin, NV20TCL_ALPHA_FUNC_ENABLE, 1);
+	OUT_RING  (0);
+	BEGIN_RING(kelvin, NV20TCL_ALPHA_FUNC_FUNC, 2);
+	OUT_RING  (NV20TCL_ALPHA_FUNC_FUNC_ALWAYS);
+	OUT_RING  (0);			/* NV20TCL_ALPHA_FUNC_REF */
+
+	for (i = 0; i < NV20TCL_TX_ENABLE__SIZE; ++i) {
+		BEGIN_RING(kelvin, NV20TCL_TX_ENABLE(i), 1);
+		OUT_RING  (0);
+	}
+	BEGIN_RING(kelvin, NV20TCL_TX_SHADER_OP, 1);
+	OUT_RING  (0);
+	BEGIN_RING(kelvin, NV20TCL_TX_SHADER_CULL_MODE, 1);
+	OUT_RING  (0);
+	BEGIN_RING(kelvin, NV20TCL_RC_IN_ALPHA(0), 4);
+	OUT_RING  (0x30d410d0);
+	OUT_RING  (0);
+	OUT_RING  (0);
+	OUT_RING  (0);
+	BEGIN_RING(kelvin, NV20TCL_RC_OUT_RGB(0), 4);
+	OUT_RING  (0x00000c00);
+	OUT_RING  (0);
+	OUT_RING  (0);
+	OUT_RING  (0);
+	BEGIN_RING(kelvin, NV20TCL_RC_ENABLE, 1);
+	OUT_RING  (0x00011101);
+	BEGIN_RING(kelvin, NV20TCL_RC_FINAL0, 2);
+	OUT_RING  (0x130e0300);
+	OUT_RING  (0x0c091c80);
+	BEGIN_RING(kelvin, NV20TCL_RC_OUT_ALPHA(0), 4);
+	OUT_RING  (0x00000c00);
+	OUT_RING  (0);
+	OUT_RING  (0);
+	OUT_RING  (0);
+	BEGIN_RING(kelvin, NV20TCL_RC_IN_RGB(0), 4);
+	OUT_RING  (0x20c400c0);
+	OUT_RING  (0);
+	OUT_RING  (0);
+	OUT_RING  (0);
+	BEGIN_RING(kelvin, NV20TCL_RC_COLOR0, 2);
+	OUT_RING  (0);
+	OUT_RING  (0);
+	BEGIN_RING(kelvin, NV20TCL_RC_CONSTANT_COLOR0(0), 4);
+	OUT_RING  (0x035125a0);
+	OUT_RING  (0);
+	OUT_RING  (0x40002000);
+	OUT_RING  (0);
+	BEGIN_RING(kelvin, NV20TCL_MULTISAMPLE_CONTROL, 1);
+	OUT_RING  (0xffff0000);
+
+	BEGIN_RING(kelvin, NV20TCL_BLEND_FUNC_ENABLE, 1);
+	OUT_RING  (0);
+	BEGIN_RING(kelvin, NV20TCL_DITHER_ENABLE, 1);
+	OUT_RING  (0);
+	BEGIN_RING(kelvin, NV20TCL_STENCIL_ENABLE, 1);
+	OUT_RING  (0);
+	BEGIN_RING(kelvin, NV20TCL_BLEND_FUNC_SRC, 4);
+	OUT_RING  (NV20TCL_BLEND_FUNC_SRC_ONE);
+	OUT_RING  (NV20TCL_BLEND_FUNC_DST_ZERO);
+	OUT_RING  (0);			/* NV20TCL_BLEND_COLOR */
+	OUT_RING  (NV20TCL_BLEND_EQUATION_FUNC_ADD);
+	BEGIN_RING(kelvin, NV20TCL_STENCIL_MASK, 7);
+	OUT_RING  (0xff);
+	OUT_RING  (NV20TCL_STENCIL_FUNC_FUNC_ALWAYS);
+	OUT_RING  (0);			/* NV20TCL_STENCIL_FUNC_REF */
+	OUT_RING  (0xff);		/* NV20TCL_STENCIL_FUNC_MASK */
+	OUT_RING  (NV20TCL_STENCIL_OP_FAIL_KEEP);
+	OUT_RING  (NV20TCL_STENCIL_OP_ZFAIL_KEEP);
+	OUT_RING  (NV20TCL_STENCIL_OP_ZPASS_KEEP);
+
+	BEGIN_RING(kelvin, NV20TCL_COLOR_LOGIC_OP_ENABLE, 2);
+	OUT_RING  (0);
+	OUT_RING  (NV20TCL_COLOR_LOGIC_OP_OP_COPY);
+	BEGIN_RING(kelvin, 0x17cc, 1);
+	OUT_RING  (0);
+	if (is_nv25tcl) {
+		BEGIN_RING(kelvin, 0x1d84, 1);
+		OUT_RING  (1);
+	}
+	BEGIN_RING(kelvin, NV20TCL_LIGHTING_ENABLE, 1);
+	OUT_RING  (0);
+	BEGIN_RING(kelvin, NV20TCL_LIGHT_CONTROL, 1);
+	OUT_RING  (0x00020000);
+	BEGIN_RING(kelvin, NV20TCL_SEPARATE_SPECULAR_ENABLE, 1);
+	OUT_RING  (0);
+	BEGIN_RING(kelvin, NV20TCL_LIGHT_MODEL_TWO_SIDE_ENABLE, 1);
+	OUT_RING  (0);
+	BEGIN_RING(kelvin, NV20TCL_ENABLED_LIGHTS, 1);
+	OUT_RING  (0);
+	BEGIN_RING(kelvin, NV20TCL_NORMALIZE_ENABLE, 1);
+	OUT_RING  (0);
+	BEGIN_RING(kelvin, NV20TCL_POLYGON_STIPPLE_PATTERN(0),
+					NV20TCL_POLYGON_STIPPLE_PATTERN__SIZE);
+	for (i = 0; i < NV20TCL_POLYGON_STIPPLE_PATTERN__SIZE; ++i) {
+		OUT_RING(0xffffffff);
+	}
+
+	BEGIN_RING(kelvin, NV20TCL_POLYGON_OFFSET_POINT_ENABLE, 3);
+	OUT_RING  (0);
+	OUT_RING  (0);		/* NV20TCL.POLYGON_OFFSET_LINE_ENABLE */
+	OUT_RING  (0);		/* NV20TCL.POLYGON_OFFSET_FILL_ENABLE */
+	BEGIN_RING(kelvin, NV20TCL_DEPTH_FUNC, 1);
+	OUT_RING  (NV20TCL_DEPTH_FUNC_LESS);
+	BEGIN_RING(kelvin, NV20TCL_DEPTH_WRITE_ENABLE, 1);
+	OUT_RING  (0);
+	BEGIN_RING(kelvin, NV20TCL_DEPTH_TEST_ENABLE, 1);
+	OUT_RING  (0);
+	BEGIN_RING(kelvin, NV20TCL_POLYGON_OFFSET_FACTOR, 2);
+	OUT_RINGf (0.0);
+	OUT_RINGf (0.0);	/* NV20TCL.POLYGON_OFFSET_UNITS */
+	BEGIN_RING(kelvin, NV20TCL_DEPTH_UNK17D8, 1);
+	OUT_RING  (1);
+	if (!is_nv25tcl) {
+		BEGIN_RING(kelvin, 0x1d84, 1);
+		OUT_RING  (3);
+	}
+	BEGIN_RING(kelvin, NV20TCL_POINT_SIZE, 1);
+	if (!is_nv25tcl) {
+		OUT_RING  (8);
+	} else {
+		OUT_RINGf (1.0);
+	}
+	if (!is_nv25tcl) {
+		BEGIN_RING(kelvin, NV20TCL_POINT_PARAMETERS_ENABLE, 2);
+		OUT_RING  (0);
+		OUT_RING  (0);		/* NV20TCL.POINT_SMOOTH_ENABLE */
+	} else {
+		BEGIN_RING(kelvin, NV20TCL_POINT_PARAMETERS_ENABLE, 1);
+		OUT_RING  (0);
+		BEGIN_RING(kelvin, 0x0a1c, 1);
+		OUT_RING  (0x800);
+	}
+	BEGIN_RING(kelvin, NV20TCL_LINE_WIDTH, 1);
+	OUT_RING  (8);
+	BEGIN_RING(kelvin, NV20TCL_LINE_SMOOTH_ENABLE, 1);
+	OUT_RING  (0);
+	BEGIN_RING(kelvin, NV20TCL_POLYGON_MODE_FRONT, 2);
+	OUT_RING  (NV20TCL_POLYGON_MODE_FRONT_FILL);
+	OUT_RING  (NV20TCL_POLYGON_MODE_BACK_FILL);
+	BEGIN_RING(kelvin, NV20TCL_CULL_FACE, 2);
+	OUT_RING  (NV20TCL_CULL_FACE_BACK);
+	OUT_RING  (NV20TCL_FRONT_FACE_CCW);
+	BEGIN_RING(kelvin, NV20TCL_POLYGON_SMOOTH_ENABLE, 1);
+	OUT_RING  (0);
+	BEGIN_RING(kelvin, NV20TCL_CULL_FACE_ENABLE, 1);
+	OUT_RING  (0);
+	BEGIN_RING(kelvin, NV20TCL_SHADE_MODEL, 1);
+	OUT_RING  (NV20TCL_SHADE_MODEL_SMOOTH);
+	BEGIN_RING(kelvin, NV20TCL_POLYGON_STIPPLE_ENABLE, 1);
+	OUT_RING  (0);
+	BEGIN_RING(kelvin, NV20TCL_TX_GEN_S(0), 4 * NV20TCL_TX_GEN_S__SIZE);
+	for (i=0; i < 4 * NV20TCL_TX_GEN_S__SIZE; ++i) {
+		OUT_RING(0);
+	}
+	BEGIN_RING(kelvin, NV20TCL_FOG_EQUATION_CONSTANT, 3);
+	OUT_RINGf (1.5);
+	OUT_RINGf (-0.090168);		/* NV20TCL.FOG_EQUATION_LINEAR */
+	OUT_RINGf (0.0);		/* NV20TCL.FOG_EQUATION_QUADRATIC */
+	BEGIN_RING(kelvin, NV20TCL_FOG_MODE, 2);
+	OUT_RING  (NV20TCL_FOG_MODE_EXP_2);
+	OUT_RING  (NV20TCL_FOG_COORD_DIST_COORD_FOG);
+	BEGIN_RING(kelvin, NV20TCL_FOG_ENABLE, 2);
+	OUT_RING  (0);
+	OUT_RING  (0);			/* NV20TCL.FOG_COLOR */
+	BEGIN_RING(kelvin, NV20TCL_ENGINE, 1);
+	OUT_RING  (NV20TCL_ENGINE_FIXED);
+
+	for (i = 0; i < NV20TCL_TX_MATRIX_ENABLE__SIZE; ++i) {
+		BEGIN_RING(kelvin, NV20TCL_TX_MATRIX_ENABLE(i), 1);
+		OUT_RING  (0);
+	}
+
+	BEGIN_RING(kelvin, NV20TCL_VTX_ATTR_4F_X(1), 4 * 15);
+	OUT_RINGf(1.0); OUT_RINGf(0.0); OUT_RINGf(0.0); OUT_RINGf(1.0);
+	OUT_RINGf(0.0); OUT_RINGf(0.0); OUT_RINGf(1.0); OUT_RINGf(1.0);
+	OUT_RINGf(1.0); OUT_RINGf(1.0); OUT_RINGf(1.0); OUT_RINGf(1.0);
+	for (i = 4; i < 16; ++i) {
+		OUT_RINGf(0.0); OUT_RINGf(0.0); OUT_RINGf(0.0);	OUT_RINGf(1.0);
+	}
+
+	BEGIN_RING(kelvin, NV20TCL_EDGEFLAG_ENABLE, 1);
+	OUT_RING  (1);
+	BEGIN_RING(kelvin, NV20TCL_COLOR_MASK, 1);
+	OUT_RING (0x00010101);
+	BEGIN_RING(kelvin, NV20TCL_CLEAR_VALUE, 1);
+	OUT_RING (0);
+
+	memset(projectionmatrix, 0, sizeof(projectionmatrix));
+	projectionmatrix[0*4+0] = 1.0;
+	projectionmatrix[1*4+1] = 1.0;
+	projectionmatrix[2*4+2] = 16777215.0;
+	projectionmatrix[3*4+3] = 1.0;
+	BEGIN_RING(kelvin, NV20TCL_PROJECTION_MATRIX(0), 16);
+	for (i = 0; i < 16; i++) {
+		OUT_RINGf  (projectionmatrix[i]);
+	}
+
+	BEGIN_RING(kelvin, NV20TCL_DEPTH_RANGE_NEAR, 2);
+	OUT_RINGf (0.0);
+	OUT_RINGf (16777216.0); /* [0, 1] scaled approx to [0, 2^24] */
+
+	BEGIN_RING(kelvin, NV20TCL_VIEWPORT_SCALE0_X, 4);
+	OUT_RINGf (0.0); /* x-offset, w/2 + 1.031250 */
+	OUT_RINGf (0.0); /* y-offset, h/2 + 0.030762 */
+	OUT_RINGf (0.0);
+	OUT_RINGf (16777215.0);
+
+	BEGIN_RING(kelvin, NV20TCL_VIEWPORT_SCALE1_X, 4);
+	OUT_RINGf (0.0); /* no effect?, w/2 */
+	OUT_RINGf (0.0); /* no effect?, h/2 */
+	OUT_RINGf (16777215.0 * 0.5);
+	OUT_RINGf (65535.0);
+
+	FIRE_RING (NULL);
+}
+
+static void
+nv20_set_edgeflags(struct pipe_context *pipe, const unsigned *bitfield)
+{
+}
+
+struct pipe_context *
+nv20_create(struct pipe_screen *pscreen, unsigned pctx_id)
+{
+	struct nv20_screen *screen = nv20_screen(pscreen);
+	struct pipe_winsys *ws = pscreen->winsys;
+	struct nv20_context *nv20;
+	struct nouveau_winsys *nvws = screen->nvws;
+
+	nv20 = CALLOC(1, sizeof(struct nv20_context));
+	if (!nv20)
+		return NULL;
+	nv20->screen = screen;
+	nv20->pctx_id = pctx_id;
+
+	nv20->nvws = nvws;
+
+	nv20->pipe.winsys = ws;
+	nv20->pipe.screen = pscreen;
+	nv20->pipe.destroy = nv20_destroy;
+	nv20->pipe.set_edgeflags = nv20_set_edgeflags;
+	nv20->pipe.draw_arrays = nv20_draw_arrays;
+	nv20->pipe.draw_elements = nv20_draw_elements;
+	nv20->pipe.clear = nv20_clear;
+	nv20->pipe.flush = nv20_flush;
+
+	nv20_init_surface_functions(nv20);
+	nv20_init_state_functions(nv20);
+
+	nv20->draw = draw_create();
+	assert(nv20->draw);
+	draw_set_rasterize_stage(nv20->draw, nv20_draw_vbuf_stage(nv20));
+
+	nv20_init_hwctx(nv20);
+
+	return &nv20->pipe;
+}
+
diff --git a/src/gallium/drivers/nv20/nv20_context.h b/src/gallium/drivers/nv20/nv20_context.h
new file mode 100644
index 00000000000..8ad926db20a
--- /dev/null
+++ b/src/gallium/drivers/nv20/nv20_context.h
@@ -0,0 +1,153 @@
+#ifndef __NV20_CONTEXT_H__
+#define __NV20_CONTEXT_H__
+
+#include "pipe/p_context.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_state.h"
+#include "pipe/p_compiler.h"
+
+#include "util/u_memory.h"
+#include "util/u_math.h"
+
+#include "draw/draw_vertex.h"
+
+#include "nouveau/nouveau_winsys.h"
+#include "nouveau/nouveau_gldefs.h"
+
+#define NOUVEAU_PUSH_CONTEXT(ctx)                                              \
+	struct nv20_screen *ctx = nv20->screen
+#include "nouveau/nouveau_push.h"
+
+#include "nv20_state.h"
+
+#define NOUVEAU_ERR(fmt, args...) \
+	fprintf(stderr, "%s:%d -  "fmt, __func__, __LINE__, ##args);
+#define NOUVEAU_MSG(fmt, args...) \
+	fprintf(stderr, "nouveau: "fmt, ##args);
+
+#define NV20_NEW_VERTPROG	(1 << 0)
+#define NV20_NEW_FRAGPROG	(1 << 1)
+#define NV20_NEW_VTXARRAYS	(1 << 2)
+#define NV20_NEW_BLEND		(1 << 3)
+#define NV20_NEW_BLENDCOL	(1 << 4)
+#define NV20_NEW_RAST 		(1 << 5)
+#define NV20_NEW_DSA  		(1 << 6)
+#define NV20_NEW_VIEWPORT	(1 << 7)
+#define NV20_NEW_SCISSOR	(1 << 8)
+#define NV20_NEW_FRAMEBUFFER	(1 << 9)
+
+#include "nv20_screen.h"
+
+struct nv20_context {
+	struct pipe_context pipe;
+
+	struct nouveau_winsys *nvws;
+	struct nv20_screen *screen;
+	unsigned pctx_id;
+
+	struct draw_context *draw;
+
+	uint32_t dirty;
+
+	struct nv20_sampler_state *tex_sampler[PIPE_MAX_SAMPLERS];
+	struct nv20_miptree *tex_miptree[PIPE_MAX_SAMPLERS];
+	unsigned dirty_samplers;
+	unsigned fp_samplers;
+	unsigned vp_samplers;
+
+	uint32_t rt_enable;
+	struct pipe_buffer *rt[4];
+	struct pipe_buffer *zeta;
+	uint32_t lma_offset;
+
+	struct nv20_blend_state *blend;
+	struct pipe_blend_color *blend_color;
+	struct nv20_rasterizer_state *rast;
+	struct nv20_depth_stencil_alpha_state *dsa;
+	struct pipe_viewport_state *viewport;
+	struct pipe_scissor_state *scissor;
+	struct pipe_framebuffer_state *framebuffer;
+
+	//struct pipe_buffer *constbuf[PIPE_SHADER_TYPES];
+	float *constbuf[PIPE_SHADER_TYPES][32][4];
+	unsigned constbuf_nr[PIPE_SHADER_TYPES];
+
+	struct vertex_info vertex_info;
+
+	struct {
+		struct pipe_buffer *buffer;
+		uint32_t format;
+	} tex[2];
+
+	unsigned vb_enable;
+	struct {
+		struct pipe_buffer *buffer;
+		unsigned delta;
+	} vb[16];
+
+/*	struct {
+	
+		struct nouveau_resource *exec_heap;
+		struct nouveau_resource *data_heap;
+
+		struct nv20_vertex_program *active;
+
+		struct nv20_vertex_program *current;
+	} vertprog;
+*/
+	struct {
+		struct nv20_fragment_program *active;
+
+		struct nv20_fragment_program *current;
+		struct pipe_buffer *constant_buf;
+	} fragprog;
+
+	struct pipe_vertex_buffer  vtxbuf[PIPE_MAX_ATTRIBS];
+	struct pipe_vertex_element vtxelt[PIPE_MAX_ATTRIBS];
+};
+
+static INLINE struct nv20_context *
+nv20_context(struct pipe_context *pipe)
+{
+	return (struct nv20_context *)pipe;
+}
+
+extern void nv20_init_state_functions(struct nv20_context *nv20);
+extern void nv20_init_surface_functions(struct nv20_context *nv20);
+
+extern void nv20_screen_init_miptree_functions(struct pipe_screen *pscreen);
+
+/* nv20_clear.c */
+extern void nv20_clear(struct pipe_context *pipe, struct pipe_surface *ps,
+		       unsigned clearValue);
+
+/* nv20_draw.c */
+extern struct draw_stage *nv20_draw_render_stage(struct nv20_context *nv20);
+
+/* nv20_fragprog.c */
+extern void nv20_fragprog_bind(struct nv20_context *,
+			       struct nv20_fragment_program *);
+extern void nv20_fragprog_destroy(struct nv20_context *,
+				  struct nv20_fragment_program *);
+
+/* nv20_fragtex.c */
+extern void nv20_fragtex_bind(struct nv20_context *);
+
+/* nv20_prim_vbuf.c */
+struct draw_stage *nv20_draw_vbuf_stage( struct nv20_context *nv20 );
+extern void nv20_vtxbuf_bind(struct nv20_context* nv20);
+
+/* nv20_state.c and friends */
+extern void nv20_emit_hw_state(struct nv20_context *nv20);
+extern void nv20_state_tex_update(struct nv20_context *nv20);
+
+/* nv20_vbo.c */
+extern boolean nv20_draw_arrays(struct pipe_context *, unsigned mode,
+				unsigned start, unsigned count);
+extern boolean nv20_draw_elements( struct pipe_context *pipe,
+                    struct pipe_buffer *indexBuffer,
+                    unsigned indexSize,
+                    unsigned prim, unsigned start, unsigned count);
+
+
+#endif
diff --git a/src/gallium/drivers/nv20/nv20_fragprog.c b/src/gallium/drivers/nv20/nv20_fragprog.c
new file mode 100644
index 00000000000..4f496369dd3
--- /dev/null
+++ b/src/gallium/drivers/nv20/nv20_fragprog.c
@@ -0,0 +1,21 @@
+#include "pipe/p_context.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_state.h"
+
+#include "pipe/p_shader_tokens.h"
+#include "tgsi/tgsi_parse.h"
+#include "tgsi/tgsi_util.h"
+
+#include "nv20_context.h"
+
+void
+nv20_fragprog_bind(struct nv20_context *nv20, struct nv20_fragment_program *fp)
+{
+}
+
+void
+nv20_fragprog_destroy(struct nv20_context *nv20,
+		      struct nv20_fragment_program *fp)
+{
+}
+
diff --git a/src/gallium/drivers/nv20/nv20_fragtex.c b/src/gallium/drivers/nv20/nv20_fragtex.c
new file mode 100644
index 00000000000..495a7be9127
--- /dev/null
+++ b/src/gallium/drivers/nv20/nv20_fragtex.c
@@ -0,0 +1,124 @@
+#include "nv20_context.h"
+#include "nouveau/nouveau_util.h"
+
+#define _(m,tf)                                                                \
+{                                                                              \
+  TRUE,                                                                        \
+  PIPE_FORMAT_##m,                                                             \
+  NV20TCL_TX_FORMAT_FORMAT_##tf,                                               \
+}
+
+struct nv20_texture_format {
+	boolean defined;
+	uint	pipe;
+	int     format;
+};
+
+static struct nv20_texture_format
+nv20_texture_formats[] = {
+	_(A8R8G8B8_UNORM, A8R8G8B8),
+	_(A1R5G5B5_UNORM, A1R5G5B5),
+	_(A4R4G4B4_UNORM, A4R4G4B4),
+	_(L8_UNORM      , L8      ),
+	_(A8_UNORM      , A8      ),
+	_(A8L8_UNORM    , A8L8    ),
+/*	_(RGB_DXT1      , DXT1,   ), */
+/*	_(RGBA_DXT1     , DXT1,   ), */
+/*	_(RGBA_DXT3     , DXT3,   ), */
+/*	_(RGBA_DXT5     , DXT5,   ), */
+	{},
+};
+
+static struct nv20_texture_format *
+nv20_fragtex_format(uint pipe_format)
+{
+	struct nv20_texture_format *tf = nv20_texture_formats;
+
+	while (tf->defined) {
+		if (tf->pipe == pipe_format)
+			return tf;
+		tf++;
+	}
+
+	return NULL;
+}
+
+
+static void
+nv20_fragtex_build(struct nv20_context *nv20, int unit)
+{
+#if 0
+	struct nv20_sampler_state *ps = nv20->tex_sampler[unit];
+	struct nv20_miptree *nv20mt = nv20->tex_miptree[unit];
+	struct pipe_texture *pt = &nv20mt->base;
+	struct nv20_texture_format *tf;
+	uint32_t txf, txs, txp;
+
+	tf = nv20_fragtex_format(pt->format);
+	if (!tf || !tf->defined) {
+		NOUVEAU_ERR("Unsupported texture format: 0x%x\n", pt->format);
+		return;
+	}
+
+	txf  = tf->format << 8;
+	txf |= (pt->last_level + 1) << 16;
+	txf |= log2i(pt->width[0]) << 20;
+	txf |= log2i(pt->height[0]) << 24;
+	txf |= log2i(pt->depth[0]) << 28;
+	txf |= 8;
+
+	switch (pt->target) {
+	case PIPE_TEXTURE_CUBE:
+		txf |= NV10TCL_TX_FORMAT_CUBE_MAP;
+		/* fall-through */
+	case PIPE_TEXTURE_2D:
+		txf |= (2<<4);
+		break;
+	case PIPE_TEXTURE_1D:
+		txf |= (1<<4);
+		break;
+	default:
+		NOUVEAU_ERR("Unknown target %d\n", pt->target);
+		return;
+	}
+
+	BEGIN_RING(kelvin, NV10TCL_TX_OFFSET(unit), 8);
+	OUT_RELOCl(nv20mt->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD);
+	OUT_RELOCd(nv20mt->buffer,txf,NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_OR | NOUVEAU_BO_RD, 1/*VRAM*/,2/*TT*/);
+	OUT_RING  (ps->wrap);
+	OUT_RING  (0x40000000); /* enable */
+	OUT_RING  (txs);
+	OUT_RING  (ps->filt | 0x2000 /* magic */);
+	OUT_RING  ((pt->width[0] << 16) | pt->height[0]);
+	OUT_RING  (ps->bcol);
+#endif
+}
+
+void
+nv20_fragtex_bind(struct nv20_context *nv20)
+{
+#if 0
+	struct nv20_fragment_program *fp = nv20->fragprog.active;
+	unsigned samplers, unit;
+
+	samplers = nv20->fp_samplers & ~fp->samplers;
+	while (samplers) {
+		unit = ffs(samplers) - 1;
+		samplers &= ~(1 << unit);
+
+		BEGIN_RING(kelvin, NV10TCL_TX_ENABLE(unit), 1);
+		OUT_RING  (0);
+	}
+
+	samplers = nv20->dirty_samplers & fp->samplers;
+	while (samplers) {
+		unit = ffs(samplers) - 1;
+		samplers &= ~(1 << unit);
+
+		nv20_fragtex_build(nv20, unit);
+	}
+
+	nv20->fp_samplers = fp->samplers;
+#endif
+}
+
diff --git a/src/gallium/drivers/nv20/nv20_miptree.c b/src/gallium/drivers/nv20/nv20_miptree.c
new file mode 100644
index 00000000000..ef7e9c5428e
--- /dev/null
+++ b/src/gallium/drivers/nv20/nv20_miptree.c
@@ -0,0 +1,206 @@
+#include "pipe/p_state.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_inlines.h"
+
+#include "nv20_context.h"
+#include "nv20_screen.h"
+
+static void
+nv20_miptree_layout(struct nv20_miptree *nv20mt)
+{
+	struct pipe_texture *pt = &nv20mt->base;
+	boolean swizzled = FALSE;
+	uint width = pt->width[0], height = pt->height[0];
+	uint offset = 0;
+	int nr_faces, l, f;
+
+	if (pt->target == PIPE_TEXTURE_CUBE) {
+		nr_faces = 6;
+	} else {
+		nr_faces = 1;
+	}
+	
+	for (l = 0; l <= pt->last_level; l++) {
+		pt->width[l] = width;
+		pt->height[l] = height;
+		pt->nblocksx[l] = pf_get_nblocksx(&pt->block, width);
+		pt->nblocksy[l] = pf_get_nblocksy(&pt->block, height);
+
+		if (swizzled)
+			nv20mt->level[l].pitch = pt->nblocksx[l] * pt->block.size;
+		else
+			nv20mt->level[l].pitch = pt->nblocksx[0] * pt->block.size;
+		nv20mt->level[l].pitch = (nv20mt->level[l].pitch + 63) & ~63;
+
+		nv20mt->level[l].image_offset =
+			CALLOC(nr_faces, sizeof(unsigned));
+
+		width  = MAX2(1, width  >> 1);
+		height = MAX2(1, height >> 1);
+
+	}
+
+	for (f = 0; f < nr_faces; f++) {
+		for (l = 0; l <= pt->last_level; l++) {
+			nv20mt->level[l].image_offset[f] = offset;
+			offset += nv20mt->level[l].pitch * pt->height[l];
+		}
+	}
+
+	nv20mt->total_size = offset;
+}
+
+static struct pipe_texture *
+nv20_miptree_blanket(struct pipe_screen *pscreen, const struct pipe_texture *pt,
+		     const unsigned *stride, struct pipe_buffer *pb)
+{
+	struct nv20_miptree *mt;
+
+	/* Only supports 2D, non-mipmapped textures for the moment */
+	if (pt->target != PIPE_TEXTURE_2D || pt->last_level != 0 ||
+	    pt->depth[0] != 1)
+		return NULL;
+
+	mt = CALLOC_STRUCT(nv20_miptree);
+	if (!mt)
+		return NULL;
+
+	mt->base = *pt;
+	mt->base.refcount = 1;
+	mt->base.screen = pscreen;
+	mt->level[0].pitch = stride[0];
+	mt->level[0].image_offset = CALLOC(1, sizeof(unsigned));
+
+	pipe_buffer_reference(pscreen, &mt->buffer, pb);
+	return &mt->base;
+}
+
+static struct pipe_texture *
+nv20_miptree_create(struct pipe_screen *screen, const struct pipe_texture *pt)
+{
+	struct pipe_winsys *ws = screen->winsys;
+	struct nv20_miptree *mt;
+	unsigned buf_usage = PIPE_BUFFER_USAGE_PIXEL |
+	                     NOUVEAU_BUFFER_USAGE_TEXTURE;
+
+	mt = MALLOC(sizeof(struct nv20_miptree));
+	if (!mt)
+		return NULL;
+	mt->base = *pt;
+	mt->base.refcount = 1;
+	mt->base.screen = screen;
+
+	/* Swizzled textures must be POT */
+	if (pt->width[0] & (pt->width[0] - 1) ||
+	    pt->height[0] & (pt->height[0] - 1))
+		mt->base.tex_usage |= NOUVEAU_TEXTURE_USAGE_LINEAR;
+	else
+	if (pt->tex_usage & (PIPE_TEXTURE_USAGE_PRIMARY |
+	                     PIPE_TEXTURE_USAGE_DISPLAY_TARGET))
+		mt->base.tex_usage |= NOUVEAU_TEXTURE_USAGE_LINEAR;
+	else
+	if (pt->tex_usage & PIPE_TEXTURE_USAGE_DYNAMIC)
+		mt->base.tex_usage |= NOUVEAU_TEXTURE_USAGE_LINEAR;
+	else {
+		switch (pt->format) {
+		/* TODO: Figure out which formats can be swizzled */
+		case PIPE_FORMAT_A8R8G8B8_UNORM:
+		case PIPE_FORMAT_X8R8G8B8_UNORM:
+		case PIPE_FORMAT_R16_SNORM:
+			break;
+		default:
+			mt->base.tex_usage |= NOUVEAU_TEXTURE_USAGE_LINEAR;
+		}
+	}
+
+	if (pt->tex_usage & PIPE_TEXTURE_USAGE_DYNAMIC)
+		buf_usage |= PIPE_BUFFER_USAGE_CPU_READ_WRITE;
+
+	nv20_miptree_layout(mt);
+
+	mt->buffer = ws->buffer_create(ws, 256, buf_usage, mt->total_size);
+	if (!mt->buffer) {
+		FREE(mt);
+		return NULL;
+	}
+	
+	return &mt->base;
+}
+
+static void
+nv20_miptree_release(struct pipe_screen *screen, struct pipe_texture **pt)
+{
+	struct pipe_texture *mt = *pt;
+
+	*pt = NULL;
+	if (--mt->refcount <= 0) {
+		struct nv20_miptree *nv20mt = (struct nv20_miptree *)mt;
+		int l;
+
+		pipe_buffer_reference(screen, &nv20mt->buffer, NULL);
+		for (l = 0; l <= mt->last_level; l++) {
+			if (nv20mt->level[l].image_offset)
+				FREE(nv20mt->level[l].image_offset);
+		}
+		FREE(nv20mt);
+	}
+}
+
+static struct pipe_surface *
+nv20_miptree_surface_get(struct pipe_screen *screen, struct pipe_texture *pt,
+			 unsigned face, unsigned level, unsigned zslice,
+			 unsigned flags)
+{
+	struct nv20_miptree *nv20mt = (struct nv20_miptree *)pt;
+	struct pipe_surface *ps;
+
+	ps = CALLOC_STRUCT(pipe_surface);
+	if (!ps)
+		return NULL;
+	pipe_texture_reference(&ps->texture, pt);
+	ps->format = pt->format;
+	ps->width = pt->width[level];
+	ps->height = pt->height[level];
+	ps->block = pt->block;
+	ps->nblocksx = pt->nblocksx[level];
+	ps->nblocksy = pt->nblocksy[level];
+	ps->stride = nv20mt->level[level].pitch;
+	ps->usage = flags;
+	ps->status = PIPE_SURFACE_STATUS_DEFINED;
+	ps->refcount = 1;
+
+	if (pt->target == PIPE_TEXTURE_CUBE) {
+		ps->offset = nv20mt->level[level].image_offset[face];
+	} else
+	if (pt->target == PIPE_TEXTURE_3D) {
+		ps->offset = nv20mt->level[level].image_offset[zslice];
+	} else {
+		ps->offset = nv20mt->level[level].image_offset[0];
+	}
+
+	return ps;
+}
+
+static void
+nv20_miptree_surface_release(struct pipe_screen *pscreen,
+			     struct pipe_surface **psurface)
+{
+	struct pipe_surface *ps = *psurface;
+
+	*psurface = NULL;
+	if (--ps->refcount > 0)
+		return;
+
+	pipe_texture_reference(&ps->texture, NULL);
+	FREE(ps);
+}
+
+void nv20_screen_init_miptree_functions(struct pipe_screen *pscreen)
+{
+	pscreen->texture_create = nv20_miptree_create;
+	pscreen->texture_blanket = nv20_miptree_blanket;
+	pscreen->texture_release = nv20_miptree_release;
+	pscreen->get_tex_surface = nv20_miptree_surface_get;
+	pscreen->tex_surface_release = nv20_miptree_surface_release;
+}
+
diff --git a/src/gallium/drivers/nv20/nv20_prim_vbuf.c b/src/gallium/drivers/nv20/nv20_prim_vbuf.c
new file mode 100644
index 00000000000..4dd7052814e
--- /dev/null
+++ b/src/gallium/drivers/nv20/nv20_prim_vbuf.c
@@ -0,0 +1,402 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * 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, sub license, 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 NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS 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.
+ * 
+ **************************************************************************/
+
+/**
+ * \file
+ * Build post-transformation, post-clipping vertex buffers and element
+ * lists by hooking into the end of the primitive pipeline and
+ * manipulating the vertex_id field in the vertex headers.
+ *
+ * XXX: work in progress 
+ * 
+ * \author José Fonseca <[email protected]>
+ * \author Keith Whitwell <[email protected]>
+ */
+
+
+#include "pipe/p_debug.h"
+#include "pipe/p_inlines.h"
+#include "pipe/internal/p_winsys_screen.h"
+
+#include "nv20_context.h"
+#include "nv20_state.h"
+
+#include "draw/draw_vbuf.h"
+
+/**
+ * Primitive renderer for nv20.
+ */
+struct nv20_vbuf_render {
+	struct vbuf_render base;
+
+	struct nv20_context *nv20;   
+
+	/** Vertex buffer in VRAM */
+	struct pipe_buffer *pbuffer;
+
+	/** Vertex buffer in normal memory */
+	void *mbuffer;
+
+	/** Vertex size in bytes */
+	/*unsigned vertex_size;*/
+
+	/** Hardware primitive */
+	unsigned hwprim;
+};
+
+/**
+ * Basically a cast wrapper.
+ */
+static INLINE struct nv20_vbuf_render *
+nv20_vbuf_render(struct vbuf_render *render)
+{
+	assert(render);
+	return (struct nv20_vbuf_render *)render;
+}
+
+void nv20_vtxbuf_bind( struct nv20_context* nv20 )
+{
+#if 0
+	int i;
+	for(i = 0; i < NV20TCL_VTXBUF_ADDRESS__SIZE; i++) {
+		BEGIN_RING(kelvin, NV20TCL_VTXBUF_ADDRESS(i), 1);
+		OUT_RING(0/*nv20->vtxbuf*/);
+		BEGIN_RING(kelvin, NV20TCL_VTXFMT(i) ,1);
+		OUT_RING(0/*XXX*/);
+	}
+#endif
+}
+
+static const struct vertex_info *
+nv20_vbuf_render_get_vertex_info( struct vbuf_render *render )
+{
+	struct nv20_vbuf_render *nv20_render = nv20_vbuf_render(render);
+	struct nv20_context *nv20 = nv20_render->nv20;
+
+	nv20_emit_hw_state(nv20);
+
+	return &nv20->vertex_info;
+}
+
+static void *
+nv20__allocate_mbuffer(struct nv20_vbuf_render *nv20_render, size_t size)
+{
+	nv20_render->mbuffer = MALLOC(size);
+	return nv20_render->mbuffer;
+}
+
+static void *
+nv20__allocate_pbuffer(struct nv20_vbuf_render *nv20_render, size_t size)
+{
+	struct pipe_winsys *winsys = nv20_render->nv20->pipe.winsys;
+	nv20_render->pbuffer = winsys->buffer_create(winsys, 64,
+					PIPE_BUFFER_USAGE_VERTEX, size);
+	return winsys->buffer_map(winsys,
+			nv20_render->pbuffer,
+			PIPE_BUFFER_USAGE_CPU_WRITE);
+}
+
+static void *
+nv20_vbuf_render_allocate_vertices( struct vbuf_render *render,
+		ushort vertex_size,
+		ushort nr_vertices )
+{
+	struct nv20_vbuf_render *nv20_render = nv20_vbuf_render(render);
+	size_t size = (size_t)vertex_size * (size_t)nr_vertices;
+	void *buf;
+
+	assert(!nv20_render->pbuffer);
+	assert(!nv20_render->mbuffer);
+
+	/*
+	 * For small amount of vertices, don't bother with pipe vertex
+	 * buffer, the data will be passed directly via the fifo.
+	 */
+	/* XXX: Pipe vertex buffers don't work. */
+	if (0 && size > 16 * 1024)
+		buf = nv20__allocate_pbuffer(nv20_render, size);
+	else
+		buf = nv20__allocate_mbuffer(nv20_render, size);
+
+	if (buf)
+		nv20_render->nv20->dirty |= NV20_NEW_VTXARRAYS;
+
+	return buf;
+}
+
+static boolean
+nv20_vbuf_render_set_primitive( struct vbuf_render *render, 
+		unsigned prim )
+{
+	struct nv20_vbuf_render *nv20_render = nv20_vbuf_render(render);
+	unsigned hwp = nvgl_primitive(prim);
+	if (hwp == 0)
+		return FALSE;
+
+	nv20_render->hwprim = hwp;
+	return TRUE;
+}
+
+static uint32_t
+nv20__vtxhwformat(unsigned stride, unsigned fields, unsigned type)
+{
+	return (stride << NV20TCL_VTXFMT_STRIDE_SHIFT) |
+		(fields << NV20TCL_VTXFMT_SIZE_SHIFT) |
+		(type << NV20TCL_VTXFMT_TYPE_SHIFT);
+}
+
+static unsigned
+nv20__emit_format(struct nv20_context *nv20, enum attrib_emit type, int hwattr)
+{
+	uint32_t hwfmt = 0;
+	unsigned fields;
+
+	switch (type) {
+	case EMIT_OMIT:
+		hwfmt = nv20__vtxhwformat(0, 0, 2);
+		fields = 0;
+		break;
+	case EMIT_1F:
+		hwfmt = nv20__vtxhwformat(4, 1, 2);
+		fields = 1;
+		break;
+	case EMIT_2F:
+		hwfmt = nv20__vtxhwformat(8, 2, 2);
+		fields = 2;
+		break;
+	case EMIT_3F:
+		hwfmt = nv20__vtxhwformat(12, 3, 2);
+		fields = 3;
+		break;
+	case EMIT_4F:
+		hwfmt = nv20__vtxhwformat(16, 4, 2);
+		fields = 4;
+		break;
+	default:
+		NOUVEAU_ERR("unhandled attrib_emit %d\n", type);
+		return 0;
+	}
+
+	BEGIN_RING(kelvin, NV20TCL_VTXFMT(hwattr), 1);
+	OUT_RING(hwfmt);
+	return fields;
+}
+
+static unsigned
+nv20__emit_vertex_array_format(struct nv20_context *nv20)
+{
+	struct vertex_info *vinfo = &nv20->vertex_info;
+	int hwattr = NV20TCL_VTXFMT__SIZE;
+	int attr = 0;
+	unsigned nr_fields = 0;
+
+	while (hwattr-- > 0) {
+		if (vinfo->hwfmt[0] & (1 << hwattr)) {
+			nr_fields += nv20__emit_format(nv20,
+					vinfo->attrib[attr].emit, hwattr);
+			attr++;
+		} else
+			nv20__emit_format(nv20, EMIT_OMIT, hwattr);
+	}
+
+	return nr_fields;
+}
+
+static void
+nv20__draw_mbuffer(struct nv20_vbuf_render *nv20_render,
+		const ushort *indices,
+		uint nr_indices)
+{
+	struct nv20_context *nv20 = nv20_render->nv20;
+	struct vertex_info *vinfo = &nv20->vertex_info;
+	unsigned nr_fields;
+	int max_push;
+	ubyte *data = nv20_render->mbuffer;
+	int vsz = 4 * vinfo->size;
+
+	nr_fields = nv20__emit_vertex_array_format(nv20);
+
+	BEGIN_RING(kelvin, NV20TCL_VERTEX_BEGIN_END, 1);
+	OUT_RING(nv20_render->hwprim);
+
+	max_push = 1200 / nr_fields;
+	while (nr_indices) {
+		int i;
+		int push = MIN2(nr_indices, max_push);
+
+		BEGIN_RING_NI(kelvin, NV20TCL_VERTEX_DATA, push * nr_fields);
+		for (i = 0; i < push; i++) {
+			/* XXX: fixme to handle other than floats? */
+			int f = nr_fields;
+			float *attrv = (float*)&data[indices[i] * vsz];
+			while (f-- > 0)
+				OUT_RINGf(*attrv++);
+		}
+
+		nr_indices -= push;
+		indices += push;
+	}
+
+	BEGIN_RING(kelvin, NV20TCL_VERTEX_BEGIN_END, 1);
+	OUT_RING(NV20TCL_VERTEX_BEGIN_END_STOP);
+}
+
+static void
+nv20__draw_pbuffer(struct nv20_vbuf_render *nv20_render,
+		const ushort *indices,
+		uint nr_indices)
+{
+	struct nv20_context *nv20 = nv20_render->nv20;
+	int push, i;
+
+	NOUVEAU_ERR("nv20__draw_pbuffer: this path is broken.\n");
+
+	BEGIN_RING(kelvin, NV10TCL_VERTEX_ARRAY_OFFSET_POS, 1);
+	OUT_RELOCl(nv20_render->pbuffer, 0,
+			NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD);
+
+	BEGIN_RING(kelvin, NV10TCL_VERTEX_BUFFER_BEGIN_END, 1);
+	OUT_RING(nv20_render->hwprim);
+
+	if (nr_indices & 1) {
+		BEGIN_RING(kelvin, NV10TCL_VB_ELEMENT_U32, 1);
+		OUT_RING  (indices[0]);
+		indices++; nr_indices--;
+	}
+
+	while (nr_indices) {
+		// XXX too big/small ? check the size
+		push = MIN2(nr_indices, 1200 * 2);
+
+		BEGIN_RING_NI(kelvin, NV10TCL_VB_ELEMENT_U16, push >> 1);
+		for (i = 0; i < push; i+=2)
+			OUT_RING((indices[i+1] << 16) | indices[i]);
+
+		nr_indices -= push;
+		indices  += push;
+	}
+
+	BEGIN_RING(kelvin, NV10TCL_VERTEX_BUFFER_BEGIN_END, 1);
+	OUT_RING  (0);
+}
+
+static void
+nv20_vbuf_render_draw( struct vbuf_render *render,
+		const ushort *indices,
+		uint nr_indices)
+{
+	struct nv20_vbuf_render *nv20_render = nv20_vbuf_render(render);
+
+	nv20_emit_hw_state(nv20_render->nv20);
+
+	if (nv20_render->pbuffer)
+		nv20__draw_pbuffer(nv20_render, indices, nr_indices);
+	else if (nv20_render->mbuffer)
+		nv20__draw_mbuffer(nv20_render, indices, nr_indices);
+	else
+		assert(0);
+}
+
+
+static void
+nv20_vbuf_render_release_vertices( struct vbuf_render *render,
+		void *vertices, 
+		unsigned vertex_size,
+		unsigned vertices_used )
+{
+	struct nv20_vbuf_render *nv20_render = nv20_vbuf_render(render);
+	struct nv20_context *nv20 = nv20_render->nv20;
+	struct pipe_winsys *winsys = nv20->pipe.winsys;
+	struct pipe_screen *pscreen = &nv20->screen->pipe;
+
+	if (nv20_render->pbuffer) {
+		winsys->buffer_unmap(winsys, nv20_render->pbuffer);
+		pipe_buffer_reference(pscreen, &nv20_render->pbuffer, NULL);
+	} else if (nv20_render->mbuffer) {
+		FREE(nv20_render->mbuffer);
+		nv20_render->mbuffer = NULL;
+	} else
+		assert(0);
+}
+
+
+static void
+nv20_vbuf_render_destroy( struct vbuf_render *render )
+{
+	struct nv20_vbuf_render *nv20_render = nv20_vbuf_render(render);
+
+	assert(!nv20_render->pbuffer);
+	assert(!nv20_render->mbuffer);
+
+	FREE(nv20_render);
+}
+
+
+/**
+ * Create a new primitive render.
+ */
+static struct vbuf_render *
+nv20_vbuf_render_create( struct nv20_context *nv20 )
+{
+	struct nv20_vbuf_render *nv20_render = CALLOC_STRUCT(nv20_vbuf_render);
+
+	nv20_render->nv20 = nv20;
+
+	nv20_render->base.max_vertex_buffer_bytes = 16*1024;
+	nv20_render->base.max_indices = 1024;
+	nv20_render->base.get_vertex_info = nv20_vbuf_render_get_vertex_info;
+	nv20_render->base.allocate_vertices =
+					nv20_vbuf_render_allocate_vertices;
+	nv20_render->base.set_primitive = nv20_vbuf_render_set_primitive;
+	nv20_render->base.draw = nv20_vbuf_render_draw;
+	nv20_render->base.release_vertices = nv20_vbuf_render_release_vertices;
+	nv20_render->base.destroy = nv20_vbuf_render_destroy;
+
+	return &nv20_render->base;
+}
+
+
+/**
+ * Create a new primitive vbuf/render stage.
+ */
+struct draw_stage *nv20_draw_vbuf_stage( struct nv20_context *nv20 )
+{
+	struct vbuf_render *render;
+	struct draw_stage *stage;
+
+	render = nv20_vbuf_render_create(nv20);
+	if(!render)
+		return NULL;
+
+	stage = draw_vbuf_stage( nv20->draw, render );
+	if(!stage) {
+		render->destroy(render);
+		return NULL;
+	}
+
+	return stage;
+}
diff --git a/src/gallium/drivers/nv20/nv20_screen.c b/src/gallium/drivers/nv20/nv20_screen.c
new file mode 100644
index 00000000000..5f2b7b4f71f
--- /dev/null
+++ b/src/gallium/drivers/nv20/nv20_screen.c
@@ -0,0 +1,222 @@
+#include "pipe/p_screen.h"
+#include "util/u_simple_screen.h"
+
+#include "nv20_context.h"
+#include "nv20_screen.h"
+
+static const char *
+nv20_screen_get_name(struct pipe_screen *screen)
+{
+	struct nv20_screen *nv20screen = nv20_screen(screen);
+	struct nouveau_device *dev = nv20screen->nvws->channel->device;
+	static char buffer[128];
+
+	snprintf(buffer, sizeof(buffer), "NV%02X", dev->chipset);
+	return buffer;
+}
+
+static const char *
+nv20_screen_get_vendor(struct pipe_screen *screen)
+{
+	return "nouveau";
+}
+
+static int
+nv20_screen_get_param(struct pipe_screen *screen, int param)
+{
+	switch (param) {
+	case PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS:
+		return 2;
+	case PIPE_CAP_NPOT_TEXTURES:
+		return 0;
+	case PIPE_CAP_TWO_SIDED_STENCIL:
+		return 0;
+	case PIPE_CAP_GLSL:
+		return 0;
+	case PIPE_CAP_S3TC:
+		return 0;
+	case PIPE_CAP_ANISOTROPIC_FILTER:
+		return 1;
+	case PIPE_CAP_POINT_SPRITE:
+		return 0;
+	case PIPE_CAP_MAX_RENDER_TARGETS:
+		return 1;
+	case PIPE_CAP_OCCLUSION_QUERY:
+		return 0;
+	case PIPE_CAP_TEXTURE_SHADOW_MAP:
+		return 0;
+	case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
+		return 12;
+	case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
+		return 0;
+	case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
+		return 12;
+	case PIPE_CAP_MAX_VERTEX_TEXTURE_UNITS:
+		return 0;
+	case NOUVEAU_CAP_HW_VTXBUF:
+	case NOUVEAU_CAP_HW_IDXBUF:
+		return 0;
+	default:
+		NOUVEAU_ERR("Unknown PIPE_CAP %d\n", param);
+		return 0;
+	}
+}
+
+static float
+nv20_screen_get_paramf(struct pipe_screen *screen, int param)
+{
+	switch (param) {
+	case PIPE_CAP_MAX_LINE_WIDTH:
+	case PIPE_CAP_MAX_LINE_WIDTH_AA:
+		return 10.0;
+	case PIPE_CAP_MAX_POINT_WIDTH:
+	case PIPE_CAP_MAX_POINT_WIDTH_AA:
+		return 64.0;
+	case PIPE_CAP_MAX_TEXTURE_ANISOTROPY:
+		return 2.0;
+	case PIPE_CAP_MAX_TEXTURE_LOD_BIAS:
+		return 4.0;
+	default:
+		NOUVEAU_ERR("Unknown PIPE_CAP %d\n", param);
+		return 0.0;
+	}
+}
+
+static boolean
+nv20_screen_is_format_supported(struct pipe_screen *screen,
+				enum pipe_format format,
+				enum pipe_texture_target target,
+				unsigned tex_usage, unsigned geom_flags)
+{
+	if (tex_usage & PIPE_TEXTURE_USAGE_RENDER_TARGET) {
+		switch (format) {
+		case PIPE_FORMAT_A8R8G8B8_UNORM:
+		case PIPE_FORMAT_R5G6B5_UNORM: 
+		case PIPE_FORMAT_Z24S8_UNORM:
+		case PIPE_FORMAT_Z16_UNORM:
+			return TRUE;
+		default:
+			break;
+		}
+	} else {
+		switch (format) {
+		case PIPE_FORMAT_A8R8G8B8_UNORM:
+		case PIPE_FORMAT_A1R5G5B5_UNORM:
+		case PIPE_FORMAT_A4R4G4B4_UNORM:
+		case PIPE_FORMAT_R5G6B5_UNORM: 
+		case PIPE_FORMAT_L8_UNORM:
+		case PIPE_FORMAT_A8_UNORM:
+		case PIPE_FORMAT_I8_UNORM:
+			return TRUE;
+		default:
+			break;
+		}
+	}
+
+	return FALSE;
+}
+
+static void *
+nv20_surface_map(struct pipe_screen *screen, struct pipe_surface *surface,
+		 unsigned flags )
+{
+	struct pipe_winsys *ws = screen->winsys;
+	void *map;
+	struct nv20_miptree *nv20mt = (struct nv20_miptree *)surface->texture;
+
+	map = ws->buffer_map(ws, nv20mt->buffer, flags);
+	if (!map)
+		return NULL;
+
+	return map + surface->offset;
+}
+
+static void
+nv20_surface_unmap(struct pipe_screen *screen, struct pipe_surface *surface)
+{
+	struct pipe_winsys *ws = screen->winsys;
+	struct nv20_miptree *nv20mt = (struct nv20_miptree *)surface->texture;
+
+	ws->buffer_unmap(ws, nv20mt->buffer);
+}
+
+static void
+nv20_screen_destroy(struct pipe_screen *pscreen)
+{
+	struct nv20_screen *screen = nv20_screen(pscreen);
+	struct nouveau_winsys *nvws = screen->nvws;
+
+	nvws->notifier_free(&screen->sync);
+	nvws->grobj_free(&screen->kelvin);
+
+	FREE(pscreen);
+}
+
+static struct pipe_buffer *
+nv20_surface_buffer(struct pipe_surface *surf)
+{
+	struct nv20_miptree *mt = (struct nv20_miptree *)surf->texture;
+
+	return mt->buffer;
+}
+
+struct pipe_screen *
+nv20_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws)
+{
+	struct nv20_screen *screen = CALLOC_STRUCT(nv20_screen);
+	unsigned kelvin_class = 0;
+	unsigned chipset = nvws->channel->device->chipset;
+	int ret;
+
+	if (!screen)
+		return NULL;
+	screen->nvws = nvws;
+
+	/* 2D engine setup */
+	screen->eng2d = nv04_surface_2d_init(nvws);
+	screen->eng2d->buf = nv20_surface_buffer;
+
+	/* 3D object */
+	if (chipset >= 0x25)
+		kelvin_class = NV25TCL;
+	else if (chipset >= 0x20)
+		kelvin_class = NV20TCL;
+
+	if (!kelvin_class || chipset >= 0x30) {
+		NOUVEAU_ERR("Unknown nv2x chipset: nv%02x\n", chipset);
+		return NULL;
+	}
+
+	ret = nvws->grobj_alloc(nvws, kelvin_class, &screen->kelvin);
+	if (ret) {
+		NOUVEAU_ERR("Error creating 3D object: %d\n", ret);
+		return FALSE;
+	}
+
+	/* Notifier for sync purposes */
+	ret = nvws->notifier_alloc(nvws, 1, &screen->sync);
+	if (ret) {
+		NOUVEAU_ERR("Error creating notifier object: %d\n", ret);
+		nv20_screen_destroy(&screen->pipe);
+		return NULL;
+	}
+
+	screen->pipe.winsys = ws;
+	screen->pipe.destroy = nv20_screen_destroy;
+
+	screen->pipe.get_name = nv20_screen_get_name;
+	screen->pipe.get_vendor = nv20_screen_get_vendor;
+	screen->pipe.get_param = nv20_screen_get_param;
+	screen->pipe.get_paramf = nv20_screen_get_paramf;
+
+	screen->pipe.is_format_supported = nv20_screen_is_format_supported;
+
+	screen->pipe.surface_map = nv20_surface_map;
+	screen->pipe.surface_unmap = nv20_surface_unmap;
+
+	nv20_screen_init_miptree_functions(&screen->pipe);
+	u_simple_screen_init(&screen->pipe);
+
+	return &screen->pipe;
+}
+
diff --git a/src/gallium/drivers/nv20/nv20_screen.h b/src/gallium/drivers/nv20/nv20_screen.h
new file mode 100644
index 00000000000..bf2f2c0d9fb
--- /dev/null
+++ b/src/gallium/drivers/nv20/nv20_screen.h
@@ -0,0 +1,24 @@
+#ifndef __NV20_SCREEN_H__
+#define __NV20_SCREEN_H__
+
+#include "pipe/p_screen.h"
+#include "nv04/nv04_surface_2d.h"
+
+struct nv20_screen {
+	struct pipe_screen pipe;
+
+	struct nouveau_winsys *nvws;
+
+	/* HW graphics objects */
+	struct nv04_surface_2d *eng2d;
+	struct nouveau_grobj *kelvin;
+	struct nouveau_notifier *sync;
+};
+
+static INLINE struct nv20_screen *
+nv20_screen(struct pipe_screen *screen)
+{
+	return (struct nv20_screen *)screen;
+}
+
+#endif
diff --git a/src/gallium/drivers/nv20/nv20_state.c b/src/gallium/drivers/nv20/nv20_state.c
new file mode 100644
index 00000000000..ecec4f49a0c
--- /dev/null
+++ b/src/gallium/drivers/nv20/nv20_state.c
@@ -0,0 +1,582 @@
+#include "draw/draw_context.h"
+#include "pipe/p_state.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_shader_tokens.h"
+
+#include "tgsi/tgsi_parse.h"
+
+#include "nv20_context.h"
+#include "nv20_state.h"
+
+static void *
+nv20_blend_state_create(struct pipe_context *pipe,
+			const struct pipe_blend_state *cso)
+{
+	struct nv20_blend_state *cb;
+
+	cb = MALLOC(sizeof(struct nv20_blend_state));
+
+	cb->b_enable = cso->blend_enable ? 1 : 0;
+	cb->b_srcfunc = ((nvgl_blend_func(cso->alpha_src_factor)<<16) |
+			 (nvgl_blend_func(cso->rgb_src_factor)));
+	cb->b_dstfunc = ((nvgl_blend_func(cso->alpha_dst_factor)<<16) |
+			 (nvgl_blend_func(cso->rgb_dst_factor)));
+
+	cb->c_mask = (((cso->colormask & PIPE_MASK_A) ? (0x01<<24) : 0) |
+		      ((cso->colormask & PIPE_MASK_R) ? (0x01<<16) : 0) |
+		      ((cso->colormask & PIPE_MASK_G) ? (0x01<< 8) : 0) |
+		      ((cso->colormask & PIPE_MASK_B) ? (0x01<< 0) : 0));
+
+	cb->d_enable = cso->dither ? 1 : 0;
+
+	return (void *)cb;
+}
+
+static void
+nv20_blend_state_bind(struct pipe_context *pipe, void *blend)
+{
+	struct nv20_context *nv20 = nv20_context(pipe);
+
+	nv20->blend = (struct nv20_blend_state*)blend;
+
+	nv20->dirty |= NV20_NEW_BLEND;
+}
+
+static void
+nv20_blend_state_delete(struct pipe_context *pipe, void *hwcso)
+{
+	FREE(hwcso);
+}
+
+
+static INLINE unsigned
+wrap_mode(unsigned wrap) {
+	unsigned ret;
+
+	switch (wrap) {
+	case PIPE_TEX_WRAP_REPEAT:
+		ret = NV20TCL_TX_WRAP_S_REPEAT;
+		break;
+	case PIPE_TEX_WRAP_MIRROR_REPEAT:
+		ret = NV20TCL_TX_WRAP_S_MIRRORED_REPEAT;
+		break;
+	case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
+		ret = NV20TCL_TX_WRAP_S_CLAMP_TO_EDGE;
+		break;
+	case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
+		ret = NV20TCL_TX_WRAP_S_CLAMP_TO_BORDER;
+		break;
+	case PIPE_TEX_WRAP_CLAMP:
+		ret = NV20TCL_TX_WRAP_S_CLAMP;
+		break;
+	case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
+	case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
+	case PIPE_TEX_WRAP_MIRROR_CLAMP:
+	default:
+		NOUVEAU_ERR("unknown wrap mode: %d\n", wrap);
+		ret = NV20TCL_TX_WRAP_S_REPEAT;
+		break;
+	}
+
+	return (ret >> NV20TCL_TX_WRAP_S_SHIFT);
+}
+
+static void *
+nv20_sampler_state_create(struct pipe_context *pipe,
+			  const struct pipe_sampler_state *cso)
+{
+	struct nv20_sampler_state *ps;
+	uint32_t filter = 0;
+
+	ps = MALLOC(sizeof(struct nv20_sampler_state));
+
+	ps->wrap = ((wrap_mode(cso->wrap_s) << NV20TCL_TX_WRAP_S_SHIFT) |
+		    (wrap_mode(cso->wrap_t) << NV20TCL_TX_WRAP_T_SHIFT));
+
+	ps->en = 0;
+	if (cso->max_anisotropy > 1.0) {
+		/* no idea, binary driver sets it, works without it.. meh.. */
+		ps->wrap |= (1 << 5);
+
+/*		if (cso->max_anisotropy >= 8.0) {
+			ps->en |= NV20TCL_TX_ENABLE_ANISO_8X;
+		} else
+		if (cso->max_anisotropy >= 4.0) {
+			ps->en |= NV20TCL_TX_ENABLE_ANISO_4X;
+		} else {
+			ps->en |= NV20TCL_TX_ENABLE_ANISO_2X;
+		}*/
+	}
+
+	switch (cso->mag_img_filter) {
+	case PIPE_TEX_FILTER_LINEAR:
+		filter |= NV20TCL_TX_FILTER_MAGNIFY_LINEAR;
+		break;
+	case PIPE_TEX_FILTER_NEAREST:
+	default:
+		filter |= NV20TCL_TX_FILTER_MAGNIFY_NEAREST;
+		break;
+	}
+
+	switch (cso->min_img_filter) {
+	case PIPE_TEX_FILTER_LINEAR:
+		switch (cso->min_mip_filter) {
+		case PIPE_TEX_MIPFILTER_NEAREST:
+			filter |=
+				NV20TCL_TX_FILTER_MINIFY_LINEAR_MIPMAP_NEAREST;
+			break;
+		case PIPE_TEX_MIPFILTER_LINEAR:
+			filter |= NV20TCL_TX_FILTER_MINIFY_LINEAR_MIPMAP_LINEAR;
+			break;
+		case PIPE_TEX_MIPFILTER_NONE:
+		default:
+			filter |= NV20TCL_TX_FILTER_MINIFY_LINEAR;
+			break;
+		}
+		break;
+	case PIPE_TEX_FILTER_NEAREST:
+	default:
+		switch (cso->min_mip_filter) {
+		case PIPE_TEX_MIPFILTER_NEAREST:
+			filter |=
+				NV20TCL_TX_FILTER_MINIFY_NEAREST_MIPMAP_NEAREST;
+		break;
+		case PIPE_TEX_MIPFILTER_LINEAR:
+			filter |=
+				NV20TCL_TX_FILTER_MINIFY_NEAREST_MIPMAP_LINEAR;
+			break;
+		case PIPE_TEX_MIPFILTER_NONE:
+		default:
+			filter |= NV20TCL_TX_FILTER_MINIFY_NEAREST;
+			break;
+		}
+		break;
+	}
+
+	ps->filt = filter;
+
+/*	if (cso->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
+		switch (cso->compare_func) {
+		case PIPE_FUNC_NEVER:
+			ps->wrap |= NV10TCL_TX_WRAP_RCOMP_NEVER;
+			break;
+		case PIPE_FUNC_GREATER:
+			ps->wrap |= NV10TCL_TX_WRAP_RCOMP_GREATER;
+			break;
+		case PIPE_FUNC_EQUAL:
+			ps->wrap |= NV10TCL_TX_WRAP_RCOMP_EQUAL;
+			break;
+		case PIPE_FUNC_GEQUAL:
+			ps->wrap |= NV10TCL_TX_WRAP_RCOMP_GEQUAL;
+			break;
+		case PIPE_FUNC_LESS:
+			ps->wrap |= NV10TCL_TX_WRAP_RCOMP_LESS;
+			break;
+		case PIPE_FUNC_NOTEQUAL:
+			ps->wrap |= NV10TCL_TX_WRAP_RCOMP_NOTEQUAL;
+			break;
+		case PIPE_FUNC_LEQUAL:
+			ps->wrap |= NV10TCL_TX_WRAP_RCOMP_LEQUAL;
+			break;
+		case PIPE_FUNC_ALWAYS:
+			ps->wrap |= NV10TCL_TX_WRAP_RCOMP_ALWAYS;
+			break;
+		default:
+			break;
+		}
+	}*/
+
+	ps->bcol = ((float_to_ubyte(cso->border_color[3]) << 24) |
+		    (float_to_ubyte(cso->border_color[0]) << 16) |
+		    (float_to_ubyte(cso->border_color[1]) <<  8) |
+		    (float_to_ubyte(cso->border_color[2]) <<  0));
+
+	return (void *)ps;
+}
+
+static void
+nv20_sampler_state_bind(struct pipe_context *pipe, unsigned nr, void **sampler)
+{
+	struct nv20_context *nv20 = nv20_context(pipe);
+	unsigned unit;
+
+	for (unit = 0; unit < nr; unit++) {
+		nv20->tex_sampler[unit] = sampler[unit];
+		nv20->dirty_samplers |= (1 << unit);
+	}
+}
+
+static void
+nv20_sampler_state_delete(struct pipe_context *pipe, void *hwcso)
+{
+	FREE(hwcso);
+}
+
+static void
+nv20_set_sampler_texture(struct pipe_context *pipe, unsigned nr,
+			 struct pipe_texture **miptree)
+{
+	struct nv20_context *nv20 = nv20_context(pipe);
+	unsigned unit;
+
+	for (unit = 0; unit < nr; unit++) {
+		nv20->tex_miptree[unit] = (struct nv20_miptree *)miptree[unit];
+		nv20->dirty_samplers |= (1 << unit);
+	}
+}
+
+static void *
+nv20_rasterizer_state_create(struct pipe_context *pipe,
+			     const struct pipe_rasterizer_state *cso)
+{
+	struct nv20_rasterizer_state *rs;
+	int i;
+
+	/*XXX: ignored:
+	 * 	light_twoside
+	 * 	offset_cw/ccw -nohw
+	 * 	scissor
+	 * 	point_smooth -nohw
+	 * 	multisample
+	 * 	offset_units / offset_scale
+	 */
+	rs = MALLOC(sizeof(struct nv20_rasterizer_state));
+
+	rs->templ = cso;
+	
+	rs->shade_model = cso->flatshade ? NV20TCL_SHADE_MODEL_FLAT :
+						NV20TCL_SHADE_MODEL_SMOOTH;
+
+	rs->line_width = (unsigned char)(cso->line_width * 8.0) & 0xff;
+	rs->line_smooth_en = cso->line_smooth ? 1 : 0;
+
+	/* XXX: nv20 and nv25 different! */
+	rs->point_size = *(uint32_t*)&cso->point_size;
+
+	rs->poly_smooth_en = cso->poly_smooth ? 1 : 0;
+
+	if (cso->front_winding == PIPE_WINDING_CCW) {
+		rs->front_face = NV20TCL_FRONT_FACE_CCW;
+		rs->poly_mode_front = nvgl_polygon_mode(cso->fill_ccw);
+		rs->poly_mode_back  = nvgl_polygon_mode(cso->fill_cw);
+	} else {
+		rs->front_face = NV20TCL_FRONT_FACE_CW;
+		rs->poly_mode_front = nvgl_polygon_mode(cso->fill_cw);
+		rs->poly_mode_back  = nvgl_polygon_mode(cso->fill_ccw);
+	}
+
+	switch (cso->cull_mode) {
+	case PIPE_WINDING_CCW:
+		rs->cull_face_en = 1;
+		if (cso->front_winding == PIPE_WINDING_CCW)
+			rs->cull_face    = NV20TCL_CULL_FACE_FRONT;
+		else
+			rs->cull_face    = NV20TCL_CULL_FACE_BACK;
+		break;
+	case PIPE_WINDING_CW:
+		rs->cull_face_en = 1;
+		if (cso->front_winding == PIPE_WINDING_CW)
+			rs->cull_face    = NV20TCL_CULL_FACE_FRONT;
+		else
+			rs->cull_face    = NV20TCL_CULL_FACE_BACK;
+		break;
+	case PIPE_WINDING_BOTH:
+		rs->cull_face_en = 1;
+		rs->cull_face    = NV20TCL_CULL_FACE_FRONT_AND_BACK;
+		break;
+	case PIPE_WINDING_NONE:
+	default:
+		rs->cull_face_en = 0;
+		rs->cull_face    = 0;
+		break;
+	}
+
+	if (cso->point_sprite) {
+		rs->point_sprite = (1 << 0);
+		for (i = 0; i < 8; i++) {
+			if (cso->sprite_coord_mode[i] != PIPE_SPRITE_COORD_NONE)
+				rs->point_sprite |= (1 << (8 + i));
+		}
+	} else {
+		rs->point_sprite = 0;
+	}
+
+	return (void *)rs;
+}
+
+static void
+nv20_rasterizer_state_bind(struct pipe_context *pipe, void *rast)
+{
+	struct nv20_context *nv20 = nv20_context(pipe);
+
+	nv20->rast = (struct nv20_rasterizer_state*)rast;
+
+	draw_set_rasterizer_state(nv20->draw, (nv20->rast ? nv20->rast->templ : NULL));
+
+	nv20->dirty |= NV20_NEW_RAST;
+}
+
+static void
+nv20_rasterizer_state_delete(struct pipe_context *pipe, void *hwcso)
+{
+	FREE(hwcso);
+}
+
+static void *
+nv20_depth_stencil_alpha_state_create(struct pipe_context *pipe,
+			const struct pipe_depth_stencil_alpha_state *cso)
+{
+	struct nv20_depth_stencil_alpha_state *hw;
+
+	hw = MALLOC(sizeof(struct nv20_depth_stencil_alpha_state));
+
+	hw->depth.func		= nvgl_comparison_op(cso->depth.func);
+	hw->depth.write_enable	= cso->depth.writemask ? 1 : 0;
+	hw->depth.test_enable	= cso->depth.enabled ? 1 : 0;
+
+	hw->stencil.enable = cso->stencil[0].enabled ? 1 : 0;
+	hw->stencil.wmask = cso->stencil[0].writemask;
+	hw->stencil.func = nvgl_comparison_op(cso->stencil[0].func);
+	hw->stencil.ref	= cso->stencil[0].ref_value;
+	hw->stencil.vmask = cso->stencil[0].valuemask;
+	hw->stencil.fail = nvgl_stencil_op(cso->stencil[0].fail_op);
+	hw->stencil.zfail = nvgl_stencil_op(cso->stencil[0].zfail_op);
+	hw->stencil.zpass = nvgl_stencil_op(cso->stencil[0].zpass_op);
+
+	hw->alpha.enabled = cso->alpha.enabled ? 1 : 0;
+	hw->alpha.func = nvgl_comparison_op(cso->alpha.func);
+	hw->alpha.ref  = float_to_ubyte(cso->alpha.ref_value);
+
+	return (void *)hw;
+}
+
+static void
+nv20_depth_stencil_alpha_state_bind(struct pipe_context *pipe, void *dsa)
+{
+	struct nv20_context *nv20 = nv20_context(pipe);
+
+	nv20->dsa = (struct nv20_depth_stencil_alpha_state*)dsa;
+
+	nv20->dirty |= NV20_NEW_DSA;
+}
+
+static void
+nv20_depth_stencil_alpha_state_delete(struct pipe_context *pipe, void *hwcso)
+{
+	FREE(hwcso);
+}
+
+static void *
+nv20_vp_state_create(struct pipe_context *pipe,
+		     const struct pipe_shader_state *templ)
+{
+	struct nv20_context *nv20 = nv20_context(pipe);
+
+	return draw_create_vertex_shader(nv20->draw, templ);
+}
+
+static void
+nv20_vp_state_bind(struct pipe_context *pipe, void *shader)
+{
+	struct nv20_context *nv20 = nv20_context(pipe);
+
+	draw_bind_vertex_shader(nv20->draw, (struct draw_vertex_shader *) shader);
+
+	nv20->dirty |= NV20_NEW_VERTPROG;
+}
+
+static void
+nv20_vp_state_delete(struct pipe_context *pipe, void *shader)
+{
+	struct nv20_context *nv20 = nv20_context(pipe);
+
+	draw_delete_vertex_shader(nv20->draw, (struct draw_vertex_shader *) shader);
+}
+
+static void *
+nv20_fp_state_create(struct pipe_context *pipe,
+		     const struct pipe_shader_state *cso)
+{
+	struct nv20_fragment_program *fp;
+
+	fp = CALLOC(1, sizeof(struct nv20_fragment_program));
+	fp->pipe.tokens = tgsi_dup_tokens(cso->tokens);
+	
+	tgsi_scan_shader(cso->tokens, &fp->info);
+
+	return (void *)fp;
+}
+
+static void
+nv20_fp_state_bind(struct pipe_context *pipe, void *hwcso)
+{
+	struct nv20_context *nv20 = nv20_context(pipe);
+	struct nv20_fragment_program *fp = hwcso;
+
+	nv20->fragprog.current = fp;
+	nv20->dirty |= NV20_NEW_FRAGPROG;
+}
+
+static void
+nv20_fp_state_delete(struct pipe_context *pipe, void *hwcso)
+{
+	struct nv20_context *nv20 = nv20_context(pipe);
+	struct nv20_fragment_program *fp = hwcso;
+
+	nv20_fragprog_destroy(nv20, fp);
+	FREE((void*)fp->pipe.tokens);
+	FREE(fp);
+}
+
+static void
+nv20_set_blend_color(struct pipe_context *pipe,
+		     const struct pipe_blend_color *bcol)
+{
+	struct nv20_context *nv20 = nv20_context(pipe);
+
+	nv20->blend_color = (struct pipe_blend_color*)bcol;
+
+	nv20->dirty |= NV20_NEW_BLENDCOL;
+}
+
+static void
+nv20_set_clip_state(struct pipe_context *pipe,
+		    const struct pipe_clip_state *clip)
+{
+	struct nv20_context *nv20 = nv20_context(pipe);
+
+	draw_set_clip_state(nv20->draw, clip);
+}
+
+static void
+nv20_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index,
+			 const struct pipe_constant_buffer *buf )
+{
+	struct nv20_context *nv20 = nv20_context(pipe);
+	struct pipe_winsys *ws = pipe->winsys;
+
+	assert(shader < PIPE_SHADER_TYPES);
+	assert(index == 0);
+
+	if (buf) {
+		void *mapped;
+		if (buf->buffer && buf->buffer->size &&
+                    (mapped = ws->buffer_map(ws, buf->buffer, PIPE_BUFFER_USAGE_CPU_READ)))
+		{
+			memcpy(nv20->constbuf[shader], mapped, buf->buffer->size);
+			nv20->constbuf_nr[shader] =
+				buf->buffer->size / (4 * sizeof(float));
+			ws->buffer_unmap(ws, buf->buffer);
+		}
+	}
+}
+
+static void
+nv20_set_framebuffer_state(struct pipe_context *pipe,
+			   const struct pipe_framebuffer_state *fb)
+{
+	struct nv20_context *nv20 = nv20_context(pipe);
+
+	nv20->framebuffer = (struct pipe_framebuffer_state*)fb;
+
+	nv20->dirty |= NV20_NEW_FRAMEBUFFER;
+}
+
+static void
+nv20_set_polygon_stipple(struct pipe_context *pipe,
+			 const struct pipe_poly_stipple *stipple)
+{
+	NOUVEAU_ERR("line stipple hahaha\n");
+}
+
+static void
+nv20_set_scissor_state(struct pipe_context *pipe,
+		       const struct pipe_scissor_state *s)
+{
+	struct nv20_context *nv20 = nv20_context(pipe);
+
+	nv20->scissor = (struct pipe_scissor_state*)s;
+
+	nv20->dirty |= NV20_NEW_SCISSOR;
+}
+
+static void
+nv20_set_viewport_state(struct pipe_context *pipe,
+			const struct pipe_viewport_state *vpt)
+{
+	struct nv20_context *nv20 = nv20_context(pipe);
+
+	nv20->viewport = (struct pipe_viewport_state*)vpt;
+
+	draw_set_viewport_state(nv20->draw, nv20->viewport);
+
+	nv20->dirty |= NV20_NEW_VIEWPORT;
+}
+
+static void
+nv20_set_vertex_buffers(struct pipe_context *pipe, unsigned count,
+			const struct pipe_vertex_buffer *vb)
+{
+	struct nv20_context *nv20 = nv20_context(pipe);
+
+	memcpy(nv20->vtxbuf, vb, sizeof(*vb) * count);
+	nv20->dirty |= NV20_NEW_VTXARRAYS;
+
+	draw_set_vertex_buffers(nv20->draw, count, vb);
+}
+
+static void
+nv20_set_vertex_elements(struct pipe_context *pipe, unsigned count,
+			 const struct pipe_vertex_element *ve)
+{
+	struct nv20_context *nv20 = nv20_context(pipe);
+
+	memcpy(nv20->vtxelt, ve, sizeof(*ve) * count);
+	nv20->dirty |= NV20_NEW_VTXARRAYS;
+
+	draw_set_vertex_elements(nv20->draw, count, ve);
+}
+
+void
+nv20_init_state_functions(struct nv20_context *nv20)
+{
+	nv20->pipe.create_blend_state = nv20_blend_state_create;
+	nv20->pipe.bind_blend_state = nv20_blend_state_bind;
+	nv20->pipe.delete_blend_state = nv20_blend_state_delete;
+
+	nv20->pipe.create_sampler_state = nv20_sampler_state_create;
+	nv20->pipe.bind_sampler_states = nv20_sampler_state_bind;
+	nv20->pipe.delete_sampler_state = nv20_sampler_state_delete;
+	nv20->pipe.set_sampler_textures = nv20_set_sampler_texture;
+
+	nv20->pipe.create_rasterizer_state = nv20_rasterizer_state_create;
+	nv20->pipe.bind_rasterizer_state = nv20_rasterizer_state_bind;
+	nv20->pipe.delete_rasterizer_state = nv20_rasterizer_state_delete;
+
+	nv20->pipe.create_depth_stencil_alpha_state =
+		nv20_depth_stencil_alpha_state_create;
+	nv20->pipe.bind_depth_stencil_alpha_state =
+		nv20_depth_stencil_alpha_state_bind;
+	nv20->pipe.delete_depth_stencil_alpha_state =
+		nv20_depth_stencil_alpha_state_delete;
+
+	nv20->pipe.create_vs_state = nv20_vp_state_create;
+	nv20->pipe.bind_vs_state = nv20_vp_state_bind;
+	nv20->pipe.delete_vs_state = nv20_vp_state_delete;
+
+	nv20->pipe.create_fs_state = nv20_fp_state_create;
+	nv20->pipe.bind_fs_state = nv20_fp_state_bind;
+	nv20->pipe.delete_fs_state = nv20_fp_state_delete;
+
+	nv20->pipe.set_blend_color = nv20_set_blend_color;
+	nv20->pipe.set_clip_state = nv20_set_clip_state;
+	nv20->pipe.set_constant_buffer = nv20_set_constant_buffer;
+	nv20->pipe.set_framebuffer_state = nv20_set_framebuffer_state;
+	nv20->pipe.set_polygon_stipple = nv20_set_polygon_stipple;
+	nv20->pipe.set_scissor_state = nv20_set_scissor_state;
+	nv20->pipe.set_viewport_state = nv20_set_viewport_state;
+
+	nv20->pipe.set_vertex_buffers = nv20_set_vertex_buffers;
+	nv20->pipe.set_vertex_elements = nv20_set_vertex_elements;
+}
+
diff --git a/src/gallium/drivers/nv20/nv20_state.h b/src/gallium/drivers/nv20/nv20_state.h
new file mode 100644
index 00000000000..34f402fdcbf
--- /dev/null
+++ b/src/gallium/drivers/nv20/nv20_state.h
@@ -0,0 +1,139 @@
+#ifndef __NV20_STATE_H__
+#define __NV20_STATE_H__
+
+#include "pipe/p_state.h"
+#include "tgsi/tgsi_scan.h"
+
+struct nv20_blend_state {
+	uint32_t b_enable;
+	uint32_t b_srcfunc;
+	uint32_t b_dstfunc;
+
+	uint32_t c_mask;
+
+	uint32_t d_enable;
+};
+
+struct nv20_sampler_state {
+	uint32_t wrap;
+	uint32_t en;
+	uint32_t filt;
+	uint32_t bcol;
+};
+
+struct nv20_rasterizer_state {
+	uint32_t shade_model;
+
+	uint32_t line_width;
+	uint32_t line_smooth_en;
+
+	uint32_t point_size;
+
+	uint32_t poly_smooth_en;
+	
+	uint32_t poly_mode_front;
+	uint32_t poly_mode_back;
+
+	uint32_t front_face;
+	uint32_t cull_face;
+	uint32_t cull_face_en;
+
+	uint32_t point_sprite;
+
+	const struct pipe_rasterizer_state *templ;
+};
+
+struct nv20_vertex_program_exec {
+	uint32_t data[4];
+	boolean has_branch_offset;
+	int const_index;
+};
+
+struct nv20_vertex_program_data {
+	int index; /* immediates == -1 */
+	float value[4];
+};
+
+struct nv20_vertex_program {
+	const struct pipe_shader_state *pipe;
+
+	boolean translated;
+	struct nv20_vertex_program_exec *insns;
+	unsigned nr_insns;
+	struct nv20_vertex_program_data *consts;
+	unsigned nr_consts;
+
+	struct nouveau_resource *exec;
+	unsigned exec_start;
+	struct nouveau_resource *data;
+	unsigned data_start;
+	unsigned data_start_min;
+
+	uint32_t ir;
+	uint32_t or;
+};
+
+struct nv20_fragment_program_data {
+	unsigned offset;
+	unsigned index;
+};
+
+struct nv20_fragment_program {
+	struct pipe_shader_state pipe;
+	struct tgsi_shader_info info;
+
+	boolean translated;
+	boolean on_hw;
+	unsigned samplers;
+
+	uint32_t *insn;
+	int       insn_len;
+
+	struct nv20_fragment_program_data *consts;
+	unsigned nr_consts;
+
+	struct pipe_buffer *buffer;
+
+	uint32_t fp_control;
+	uint32_t fp_reg_control;
+};
+
+
+struct nv20_depth_stencil_alpha_state {
+	struct {
+		uint32_t func;
+		uint32_t write_enable;
+		uint32_t test_enable;
+	} depth;
+
+	struct {
+		uint32_t enable;
+		uint32_t wmask;
+		uint32_t func;
+		uint32_t ref;
+		uint32_t vmask;
+		uint32_t fail;
+		uint32_t zfail;
+		uint32_t zpass;
+	} stencil;
+
+	struct {
+		uint32_t enabled;
+		uint32_t func;
+		uint32_t ref;
+	} alpha;
+};
+
+struct nv20_miptree {
+	struct pipe_texture base;
+
+	struct pipe_buffer *buffer;
+	uint total_size;
+
+	struct {
+		uint pitch;
+		uint *image_offset;
+	} level[PIPE_MAX_TEXTURE_LEVELS];
+};
+
+#endif
diff --git a/src/gallium/drivers/nv20/nv20_state_emit.c b/src/gallium/drivers/nv20/nv20_state_emit.c
new file mode 100644
index 00000000000..0f4df9ca317
--- /dev/null
+++ b/src/gallium/drivers/nv20/nv20_state_emit.c
@@ -0,0 +1,396 @@
+#include "nv20_context.h"
+#include "nv20_state.h"
+#include "draw/draw_context.h"
+
+static void nv20_state_emit_blend(struct nv20_context* nv20)
+{
+	struct nv20_blend_state *b = nv20->blend;
+
+	BEGIN_RING(kelvin, NV20TCL_DITHER_ENABLE, 1);
+	OUT_RING  (b->d_enable);
+
+	BEGIN_RING(kelvin, NV20TCL_BLEND_FUNC_ENABLE, 1);
+	OUT_RING  (b->b_enable);
+
+	BEGIN_RING(kelvin, NV20TCL_BLEND_FUNC_SRC, 2);
+	OUT_RING  (b->b_srcfunc);
+	OUT_RING  (b->b_dstfunc);
+
+	BEGIN_RING(kelvin, NV20TCL_COLOR_MASK, 1);
+	OUT_RING  (b->c_mask);
+}
+
+static void nv20_state_emit_blend_color(struct nv20_context* nv20)
+{
+	struct pipe_blend_color *c = nv20->blend_color;
+
+	BEGIN_RING(kelvin, NV20TCL_BLEND_COLOR, 1);
+	OUT_RING  ((float_to_ubyte(c->color[3]) << 24)|
+		   (float_to_ubyte(c->color[0]) << 16)|
+		   (float_to_ubyte(c->color[1]) << 8) |
+		   (float_to_ubyte(c->color[2]) << 0));
+}
+
+static void nv20_state_emit_rast(struct nv20_context* nv20)
+{
+	struct nv20_rasterizer_state *r = nv20->rast;
+
+	BEGIN_RING(kelvin, NV20TCL_SHADE_MODEL, 2);
+	OUT_RING  (r->shade_model);
+	OUT_RING  (r->line_width);
+
+
+	BEGIN_RING(kelvin, NV20TCL_POINT_SIZE, 1);
+	OUT_RING  (r->point_size);
+
+	BEGIN_RING(kelvin, NV20TCL_POLYGON_MODE_FRONT, 2);
+	OUT_RING  (r->poly_mode_front);
+	OUT_RING  (r->poly_mode_back);
+
+
+	BEGIN_RING(kelvin, NV20TCL_CULL_FACE, 2);
+	OUT_RING  (r->cull_face);
+	OUT_RING  (r->front_face);
+
+	BEGIN_RING(kelvin, NV20TCL_LINE_SMOOTH_ENABLE, 2);
+	OUT_RING  (r->line_smooth_en);
+	OUT_RING  (r->poly_smooth_en);
+
+	BEGIN_RING(kelvin, NV20TCL_CULL_FACE_ENABLE, 1);
+	OUT_RING  (r->cull_face_en);
+}
+
+static void nv20_state_emit_dsa(struct nv20_context* nv20)
+{
+	struct nv20_depth_stencil_alpha_state *d = nv20->dsa;
+
+	BEGIN_RING(kelvin, NV20TCL_DEPTH_FUNC, 1);
+	OUT_RING (d->depth.func);
+
+	BEGIN_RING(kelvin, NV20TCL_DEPTH_WRITE_ENABLE, 1);
+	OUT_RING (d->depth.write_enable);
+
+	BEGIN_RING(kelvin, NV20TCL_DEPTH_TEST_ENABLE, 1);
+	OUT_RING (d->depth.test_enable);
+
+	BEGIN_RING(kelvin, NV20TCL_DEPTH_UNK17D8, 1);
+	OUT_RING (1);
+
+#if 0
+	BEGIN_RING(kelvin, NV20TCL_STENCIL_ENABLE, 1);
+	OUT_RING (d->stencil.enable);
+	BEGIN_RING(kelvin, NV20TCL_STENCIL_MASK, 7);
+	OUT_RINGp ((uint32_t *)&(d->stencil.wmask), 7);
+#endif
+
+	BEGIN_RING(kelvin, NV20TCL_ALPHA_FUNC_ENABLE, 1);
+	OUT_RING (d->alpha.enabled);
+
+	BEGIN_RING(kelvin, NV20TCL_ALPHA_FUNC_FUNC, 1);
+	OUT_RING (d->alpha.func);
+
+	BEGIN_RING(kelvin, NV20TCL_ALPHA_FUNC_REF, 1);
+	OUT_RING (d->alpha.ref);
+}
+
+static void nv20_state_emit_viewport(struct nv20_context* nv20)
+{
+}
+
+static void nv20_state_emit_scissor(struct nv20_context* nv20)
+{
+	/* NV20TCL_SCISSOR_* is probably a software method */
+/*	struct pipe_scissor_state *s = nv20->scissor;
+	BEGIN_RING(kelvin, NV20TCL_SCISSOR_HORIZ, 2);
+	OUT_RING  (((s->maxx - s->minx) << 16) | s->minx);
+	OUT_RING  (((s->maxy - s->miny) << 16) | s->miny);*/
+}
+
+static void nv20_state_emit_framebuffer(struct nv20_context* nv20)
+{
+	struct pipe_framebuffer_state* fb = nv20->framebuffer;
+	struct pipe_surface *rt, *zeta = NULL;
+	uint32_t rt_format, w, h;
+	int colour_format = 0, zeta_format = 0;
+	struct nv20_miptree *nv20mt = 0;
+
+	w = fb->cbufs[0]->width;
+	h = fb->cbufs[0]->height;
+	colour_format = fb->cbufs[0]->format;
+	rt = fb->cbufs[0];
+
+	if (fb->zsbuf) {
+		if (colour_format) {
+			assert(w == fb->zsbuf->width);
+			assert(h == fb->zsbuf->height);
+		} else {
+			w = fb->zsbuf->width;
+			h = fb->zsbuf->height;
+		}
+
+		zeta_format = fb->zsbuf->format;
+		zeta = fb->zsbuf;
+	}
+
+	rt_format = NV20TCL_RT_FORMAT_TYPE_LINEAR | 0x20;
+
+	switch (colour_format) {
+	case PIPE_FORMAT_A8R8G8B8_UNORM:
+	case 0:
+		rt_format |= NV20TCL_RT_FORMAT_COLOR_A8R8G8B8;
+		break;
+	case PIPE_FORMAT_R5G6B5_UNORM:
+		rt_format |= NV20TCL_RT_FORMAT_COLOR_R5G6B5;
+		break;
+	default:
+		assert(0);
+	}
+
+	if (zeta) {
+		BEGIN_RING(kelvin, NV20TCL_RT_PITCH, 1);
+		OUT_RING  (rt->stride | (zeta->stride << 16));
+	} else {
+		BEGIN_RING(kelvin, NV20TCL_RT_PITCH, 1);
+		OUT_RING  (rt->stride | (rt->stride << 16));
+	}
+
+	nv20mt = (struct nv20_miptree *)rt->texture;
+	nv20->rt[0] = nv20mt->buffer;
+
+	if (zeta_format)
+	{
+		nv20mt = (struct nv20_miptree *)zeta->texture;
+		nv20->zeta = nv20mt->buffer;
+	}
+
+	BEGIN_RING(kelvin, NV20TCL_RT_HORIZ, 3);
+	OUT_RING  ((w << 16) | 0);
+	OUT_RING  ((h << 16) | 0); /*NV20TCL_RT_VERT */
+	OUT_RING  (rt_format); /* NV20TCL_RT_FORMAT */
+	BEGIN_RING(kelvin, NV20TCL_VIEWPORT_CLIP_HORIZ(0), 2);
+	OUT_RING  (((w - 1) << 16) | 0);
+	OUT_RING  (((h - 1) << 16) | 0);
+}
+
+static void nv20_vertex_layout(struct nv20_context *nv20)
+{
+	struct nv20_fragment_program *fp = nv20->fragprog.current;
+	struct draw_context *dc = nv20->draw;
+	int src;
+	int i;
+	struct vertex_info *vinfo = &nv20->vertex_info;
+	const enum interp_mode colorInterp = INTERP_LINEAR;
+	boolean colors[2] = { FALSE };
+	boolean generics[12] = { FALSE };
+	boolean fog = FALSE;
+
+	memset(vinfo, 0, sizeof(*vinfo));
+
+	/*
+	 * Assumed NV20 hardware vertex attribute order:
+	 * 0 position, 1 ?, 2 ?, 3 col0,
+	 * 4 col1?, 5 ?, 6 ?, 7 ?,
+	 * 8 ?, 9 tex0, 10 tex1, 11 tex2,
+	 * 12 tex3, 13 ?, 14 ?, 15 ?
+	 * unaccounted: wgh, nor, fog
+	 * There are total 16 attrs.
+	 * vinfo->hwfmt[0] has a used-bit corresponding to each of these.
+	 * relation to TGSI_SEMANTIC_*:
+	 * - POSITION: position (always used)
+	 * - COLOR: col1, col0
+	 * - GENERIC: tex3, tex2, tex1, tex0, normal, weight
+	 * - FOG: fog
+	 */
+
+	for (i = 0; i < fp->info.num_inputs; i++) {
+		int isn = fp->info.input_semantic_name[i];
+		int isi = fp->info.input_semantic_index[i];
+		switch (isn) {
+		case TGSI_SEMANTIC_POSITION:
+			break;
+		case TGSI_SEMANTIC_COLOR:
+			assert(isi < 2);
+			colors[isi] = TRUE;
+			break;
+		case TGSI_SEMANTIC_GENERIC:
+			assert(isi < 12);
+			generics[isi] = TRUE;
+			break;
+		case TGSI_SEMANTIC_FOG:
+			fog = TRUE;
+			break;
+		default:
+			assert(0 && "unknown input_semantic_name");
+		}
+	}
+
+	/* always do position */ {
+		src = draw_find_vs_output(dc, TGSI_SEMANTIC_POSITION, 0);
+		draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_LINEAR, src);
+		vinfo->hwfmt[0] |= (1 << 0);
+	}
+
+	/* two unnamed generics */
+	for (i = 4; i < 6; i++) {
+		if (!generics[i])
+			continue;
+		src = draw_find_vs_output(dc, TGSI_SEMANTIC_GENERIC, i);
+		draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE, src);
+		vinfo->hwfmt[0] |= (1 << (i - 3));
+	}
+
+	if (colors[0]) {
+		src = draw_find_vs_output(dc, TGSI_SEMANTIC_COLOR, 0);
+		draw_emit_vertex_attr(vinfo, EMIT_4F, colorInterp, src);
+		vinfo->hwfmt[0] |= (1 << 3);
+	}
+
+	if (colors[1]) {
+		src = draw_find_vs_output(dc, TGSI_SEMANTIC_COLOR, 1);
+		draw_emit_vertex_attr(vinfo, EMIT_4F, colorInterp, src);
+		vinfo->hwfmt[0] |= (1 << 4);
+	}
+
+	/* four unnamed generics */
+	for (i = 6; i < 10; i++) {
+		if (!generics[i])
+			continue;
+		src = draw_find_vs_output(dc, TGSI_SEMANTIC_GENERIC, i);
+		draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE, src);
+		vinfo->hwfmt[0] |= (1 << (i - 1));
+	}
+
+	/* tex0, tex1, tex2, tex3 */
+	for (i = 0; i < 4; i++) {
+		if (!generics[i])
+			continue;
+		src = draw_find_vs_output(dc, TGSI_SEMANTIC_GENERIC, i);
+		draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE, src);
+		vinfo->hwfmt[0] |= (1 << (i + 9));
+	}
+
+	/* two unnamed generics */
+	for (i = 10; i < 12; i++) {
+		if (!generics[i])
+			continue;
+		src = draw_find_vs_output(dc, TGSI_SEMANTIC_GENERIC, i);
+		draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE, src);
+		vinfo->hwfmt[0] |= (1 << (i + 3));
+	}
+
+	if (fog) {
+		src = draw_find_vs_output(dc, TGSI_SEMANTIC_FOG, 0);
+		draw_emit_vertex_attr(vinfo, EMIT_1F, INTERP_PERSPECTIVE, src);
+		vinfo->hwfmt[0] |= (1 << 15);
+	}
+
+	draw_compute_vertex_size(vinfo);
+}
+
+void
+nv20_emit_hw_state(struct nv20_context *nv20)
+{
+	int i;
+
+	if (nv20->dirty & NV20_NEW_VERTPROG) {
+		//nv20_vertprog_bind(nv20, nv20->vertprog.current);
+		nv20->dirty &= ~NV20_NEW_VERTPROG;
+	}
+
+	if (nv20->dirty & NV20_NEW_FRAGPROG) {
+		nv20_fragprog_bind(nv20, nv20->fragprog.current);
+		/*XXX: clear NV20_NEW_FRAGPROG if no new program uploaded */
+		nv20->dirty_samplers |= (1<<10);
+		nv20->dirty_samplers = 0;
+	}
+
+	if (nv20->dirty_samplers || (nv20->dirty & NV20_NEW_FRAGPROG)) {
+		nv20_fragtex_bind(nv20);
+		nv20->dirty &= ~NV20_NEW_FRAGPROG;
+	}
+
+	if (nv20->dirty & NV20_NEW_VTXARRAYS) {
+		nv20->dirty &= ~NV20_NEW_VTXARRAYS;
+		nv20_vertex_layout(nv20);
+		nv20_vtxbuf_bind(nv20);
+	}
+
+	if (nv20->dirty & NV20_NEW_BLEND) {
+		nv20->dirty &= ~NV20_NEW_BLEND;
+		nv20_state_emit_blend(nv20);
+	}
+
+	if (nv20->dirty & NV20_NEW_BLENDCOL) {
+		nv20->dirty &= ~NV20_NEW_BLENDCOL;
+		nv20_state_emit_blend_color(nv20);
+	}
+
+	if (nv20->dirty & NV20_NEW_RAST) {
+		nv20->dirty &= ~NV20_NEW_RAST;
+		nv20_state_emit_rast(nv20);
+	}
+
+	if (nv20->dirty & NV20_NEW_DSA) {
+		nv20->dirty &= ~NV20_NEW_DSA;
+		nv20_state_emit_dsa(nv20);
+	}
+
+ 	if (nv20->dirty & NV20_NEW_VIEWPORT) {
+		nv20->dirty &= ~NV20_NEW_VIEWPORT;
+		nv20_state_emit_viewport(nv20);
+	}
+
+ 	if (nv20->dirty & NV20_NEW_SCISSOR) {
+		nv20->dirty &= ~NV20_NEW_SCISSOR;
+		nv20_state_emit_scissor(nv20);
+	}
+
+ 	if (nv20->dirty & NV20_NEW_FRAMEBUFFER) {
+		nv20->dirty &= ~NV20_NEW_FRAMEBUFFER;
+		nv20_state_emit_framebuffer(nv20);
+	}
+
+	/* Emit relocs for every referenced buffer.
+	 * This is to ensure the bufmgr has an accurate idea of how
+	 * the buffer is used.  This isn't very efficient, but we don't
+	 * seem to take a significant performance hit.  Will be improved
+	 * at some point.  Vertex arrays are emitted by nv20_vbo.c
+	 */
+
+	/* Render target */
+	BEGIN_RING(kelvin, NV20TCL_DMA_COLOR, 1);
+	OUT_RELOCo(nv20->rt[0], NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
+	BEGIN_RING(kelvin, NV20TCL_COLOR_OFFSET, 1);
+	OUT_RELOCl(nv20->rt[0], 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
+
+	if (nv20->zeta) {
+		BEGIN_RING(kelvin, NV20TCL_DMA_ZETA, 1);
+		OUT_RELOCo(nv20->zeta, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
+		BEGIN_RING(kelvin, NV20TCL_ZETA_OFFSET, 1);
+		OUT_RELOCl(nv20->zeta, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
+		/* XXX for when we allocate LMA on nv17 */
+/*		BEGIN_RING(kelvin, NV10TCL_LMA_DEPTH_BUFFER_OFFSET, 1);
+		OUT_RELOCl(nv20->zeta + lma_offset);*/
+	}
+
+	/* Vertex buffer */
+	BEGIN_RING(kelvin, NV20TCL_DMA_VTXBUF0, 1);
+	OUT_RELOCo(nv20->rt[0], NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
+	BEGIN_RING(kelvin, NV20TCL_COLOR_OFFSET, 1);
+	OUT_RELOCl(nv20->rt[0], 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
+
+	/* Texture images */
+	for (i = 0; i < 2; i++) {
+		if (!(nv20->fp_samplers & (1 << i)))
+			continue;
+		BEGIN_RING(kelvin, NV20TCL_TX_OFFSET(i), 1);
+		OUT_RELOCl(nv20->tex[i].buffer, 0, NOUVEAU_BO_VRAM |
+			   NOUVEAU_BO_GART | NOUVEAU_BO_RD);
+		BEGIN_RING(kelvin, NV20TCL_TX_FORMAT(i), 1);
+		OUT_RELOCd(nv20->tex[i].buffer, nv20->tex[i].format,
+			   NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD |
+			   NOUVEAU_BO_OR, NV20TCL_TX_FORMAT_DMA0,
+			   NV20TCL_TX_FORMAT_DMA1);
+	}
+}
+
diff --git a/src/gallium/drivers/nv20/nv20_surface.c b/src/gallium/drivers/nv20/nv20_surface.c
new file mode 100644
index 00000000000..6cd607583cf
--- /dev/null
+++ b/src/gallium/drivers/nv20/nv20_surface.c
@@ -0,0 +1,72 @@
+
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * 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, sub license, 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 NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS 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.
+ * 
+ **************************************************************************/
+
+#include "nv20_context.h"
+#include "pipe/p_defines.h"
+#include "pipe/internal/p_winsys_screen.h"
+#include "pipe/p_inlines.h"
+#include "util/u_tile.h"
+
+static void
+nv20_surface_copy(struct pipe_context *pipe, boolean do_flip,
+		  struct pipe_surface *dest, unsigned destx, unsigned desty,
+		  struct pipe_surface *src, unsigned srcx, unsigned srcy,
+		  unsigned width, unsigned height)
+{
+	struct nv20_context *nv20 = nv20_context(pipe);
+	struct nv04_surface_2d *eng2d = nv20->screen->eng2d;
+
+	if (do_flip) {
+		desty += height;
+		while (height--) {
+			eng2d->copy(eng2d, dest, destx, desty--, src,
+				    srcx, srcy++, width, 1);
+		}
+		return;
+	}
+
+	eng2d->copy(eng2d, dest, destx, desty, src, srcx, srcy, width, height);
+}
+
+static void
+nv20_surface_fill(struct pipe_context *pipe, struct pipe_surface *dest,
+		  unsigned destx, unsigned desty, unsigned width,
+		  unsigned height, unsigned value)
+{
+	struct nv20_context *nv20 = nv20_context(pipe);
+	struct nv04_surface_2d *eng2d = nv20->screen->eng2d;
+
+	eng2d->fill(eng2d, dest, destx, desty, width, height, value);
+}
+
+void
+nv20_init_surface_functions(struct nv20_context *nv20)
+{
+	nv20->pipe.surface_copy = nv20_surface_copy;
+	nv20->pipe.surface_fill = nv20_surface_fill;
+}
diff --git a/src/gallium/drivers/nv20/nv20_vbo.c b/src/gallium/drivers/nv20/nv20_vbo.c
new file mode 100644
index 00000000000..24d8f4bef09
--- /dev/null
+++ b/src/gallium/drivers/nv20/nv20_vbo.c
@@ -0,0 +1,78 @@
+#include "draw/draw_context.h"
+#include "pipe/p_context.h"
+#include "pipe/p_state.h"
+
+#include "nv20_context.h"
+#include "nv20_state.h"
+
+#include "nouveau/nouveau_channel.h"
+#include "nouveau/nouveau_pushbuf.h"
+
+boolean nv20_draw_elements( struct pipe_context *pipe,
+                    struct pipe_buffer *indexBuffer,
+                    unsigned indexSize,
+                    unsigned prim, unsigned start, unsigned count)
+{
+	struct nv20_context *nv20 = nv20_context( pipe );
+	struct draw_context *draw = nv20->draw;
+	unsigned i;
+
+	nv20_emit_hw_state(nv20);
+
+	/*
+	 * Map vertex buffers
+	 */
+	for (i = 0; i < PIPE_MAX_ATTRIBS; i++) {
+		if (nv20->vtxbuf[i].buffer) {
+			void *buf
+				= pipe->winsys->buffer_map(pipe->winsys,
+						nv20->vtxbuf[i].buffer,
+						PIPE_BUFFER_USAGE_CPU_READ);
+			draw_set_mapped_vertex_buffer(draw, i, buf);
+		}
+	}
+	/* Map index buffer, if present */
+	if (indexBuffer) {
+		void *mapped_indexes
+			= pipe->winsys->buffer_map(pipe->winsys, indexBuffer,
+					PIPE_BUFFER_USAGE_CPU_READ);
+		draw_set_mapped_element_buffer(draw, indexSize, mapped_indexes);
+	}
+	else {
+		/* no index/element buffer */
+		draw_set_mapped_element_buffer(draw, 0, NULL);
+	}
+
+	draw_set_mapped_constant_buffer(draw,
+					nv20->constbuf[PIPE_SHADER_VERTEX],
+					nv20->constbuf_nr[PIPE_SHADER_VERTEX]);
+
+	/* draw! */
+	draw_arrays(nv20->draw, prim, start, count);
+
+	/*
+	 * unmap vertex/index buffers
+	 */
+	for (i = 0; i < PIPE_MAX_ATTRIBS; i++) {
+		if (nv20->vtxbuf[i].buffer) {
+			pipe->winsys->buffer_unmap(pipe->winsys, nv20->vtxbuf[i].buffer);
+			draw_set_mapped_vertex_buffer(draw, i, NULL);
+		}
+	}
+	if (indexBuffer) {
+		pipe->winsys->buffer_unmap(pipe->winsys, indexBuffer);
+		draw_set_mapped_element_buffer(draw, 0, NULL);
+	}
+
+	draw_flush(nv20->draw);
+	return TRUE;
+}
+
+boolean nv20_draw_arrays( struct pipe_context *pipe,
+				 unsigned prim, unsigned start, unsigned count)
+{
+	return nv20_draw_elements(pipe, NULL, 0, prim, start, count);
+}
+
+
+
diff --git a/src/gallium/drivers/nv20/nv20_vertprog.c b/src/gallium/drivers/nv20/nv20_vertprog.c
new file mode 100644
index 00000000000..a885fcd7a56
--- /dev/null
+++ b/src/gallium/drivers/nv20/nv20_vertprog.c
@@ -0,0 +1,838 @@
+#include "pipe/p_context.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_state.h"
+
+#include "pipe/p_shader_tokens.h"
+#include "tgsi/tgsi_parse.h"
+#include "tgsi/tgsi_dump.h"
+
+#include "nv20_context.h"
+#include "nv20_state.h"
+
+/* TODO (at least...):
+ *  1. Indexed consts  + ARL
+ *  2. Arb. swz/negation
+ *  3. NV_vp11, NV_vp2, NV_vp3 features
+ *       - extra arith opcodes
+ *       - branching
+ *       - texture sampling
+ *       - indexed attribs
+ *       - indexed results
+ *  4. bugs
+ */
+
+#define SWZ_X 0
+#define SWZ_Y 1
+#define SWZ_Z 2
+#define SWZ_W 3
+#define MASK_X 8
+#define MASK_Y 4
+#define MASK_Z 2
+#define MASK_W 1
+#define MASK_ALL (MASK_X|MASK_Y|MASK_Z|MASK_W)
+#define DEF_SCALE 0
+#define DEF_CTEST 0
+#include "nv20_shader.h"
+
+#define swz(s,x,y,z,w) nv20_sr_swz((s), SWZ_##x, SWZ_##y, SWZ_##z, SWZ_##w)
+#define neg(s) nv20_sr_neg((s))
+#define abs(s) nv20_sr_abs((s))
+
+struct nv20_vpc {
+	struct nv20_vertex_program *vp;
+
+	struct nv20_vertex_program_exec *vpi;
+
+	unsigned output_map[PIPE_MAX_SHADER_OUTPUTS];
+
+	int high_temp;
+	int temp_temp_count;
+
+	struct nv20_sreg *imm;
+	unsigned nr_imm;
+};
+
+static struct nv20_sreg
+temp(struct nv20_vpc *vpc)
+{
+	int idx;
+
+	idx  = vpc->temp_temp_count++;
+	idx += vpc->high_temp + 1;
+	return nv20_sr(NV30SR_TEMP, idx);
+}
+
+static struct nv20_sreg
+constant(struct nv20_vpc *vpc, int pipe, float x, float y, float z, float w)
+{
+	struct nv20_vertex_program *vp = vpc->vp;
+	struct nv20_vertex_program_data *vpd;
+	int idx;
+
+	if (pipe >= 0) {
+		for (idx = 0; idx < vp->nr_consts; idx++) {
+			if (vp->consts[idx].index == pipe)
+				return nv20_sr(NV30SR_CONST, idx);
+		}
+	}
+
+	idx = vp->nr_consts++;
+	vp->consts = realloc(vp->consts, sizeof(*vpd) * vp->nr_consts);
+	vpd = &vp->consts[idx];
+
+	vpd->index = pipe;
+	vpd->value[0] = x;
+	vpd->value[1] = y;
+	vpd->value[2] = z;
+	vpd->value[3] = w;
+	return nv20_sr(NV30SR_CONST, idx);
+}
+
+#define arith(cc,s,o,d,m,s0,s1,s2) \
+	nv20_vp_arith((cc), (s), NV30_VP_INST_##o, (d), (m), (s0), (s1), (s2))
+
+static void
+emit_src(struct nv20_vpc *vpc, uint32_t *hw, int pos, struct nv20_sreg src)
+{
+	struct nv20_vertex_program *vp = vpc->vp;
+	uint32_t sr = 0;
+
+	switch (src.type) {
+	case NV30SR_TEMP:
+		sr |= (NV30_VP_SRC_REG_TYPE_TEMP << NV30_VP_SRC_REG_TYPE_SHIFT);
+		sr |= (src.index << NV30_VP_SRC_TEMP_SRC_SHIFT);
+		break;
+	case NV30SR_INPUT:
+		sr |= (NV30_VP_SRC_REG_TYPE_INPUT <<
+		       NV30_VP_SRC_REG_TYPE_SHIFT);
+		vp->ir |= (1 << src.index);
+		hw[1] |= (src.index << NV30_VP_INST_INPUT_SRC_SHIFT);
+		break;
+	case NV30SR_CONST:
+		sr |= (NV30_VP_SRC_REG_TYPE_CONST <<
+		       NV30_VP_SRC_REG_TYPE_SHIFT);
+		assert(vpc->vpi->const_index == -1 ||
+		       vpc->vpi->const_index == src.index);
+		vpc->vpi->const_index = src.index;
+		break;
+	case NV30SR_NONE:
+		sr |= (NV30_VP_SRC_REG_TYPE_INPUT <<
+		       NV30_VP_SRC_REG_TYPE_SHIFT);
+		break;
+	default:
+		assert(0);
+	}
+
+	if (src.negate)
+		sr |= NV30_VP_SRC_NEGATE;
+
+	if (src.abs)
+		hw[0] |= (1 << (21 + pos));
+
+	sr |= ((src.swz[0] << NV30_VP_SRC_SWZ_X_SHIFT) |
+	       (src.swz[1] << NV30_VP_SRC_SWZ_Y_SHIFT) |
+	       (src.swz[2] << NV30_VP_SRC_SWZ_Z_SHIFT) |
+	       (src.swz[3] << NV30_VP_SRC_SWZ_W_SHIFT));
+
+/*
+ * |VVV|
+ * d�.�b
+ *  \u/
+ *
+ */
+
+	switch (pos) {
+	case 0:
+		hw[1] |= ((sr & NV30_VP_SRC0_HIGH_MASK) >>
+			  NV30_VP_SRC0_HIGH_SHIFT) << NV30_VP_INST_SRC0H_SHIFT;
+		hw[2] |= (sr & NV30_VP_SRC0_LOW_MASK) <<
+			  NV30_VP_INST_SRC0L_SHIFT;
+		break;
+	case 1:
+		hw[2] |= sr << NV30_VP_INST_SRC1_SHIFT;
+		break;
+	case 2:
+		hw[2] |= ((sr & NV30_VP_SRC2_HIGH_MASK) >>
+			  NV30_VP_SRC2_HIGH_SHIFT) << NV30_VP_INST_SRC2H_SHIFT;
+		hw[3] |= (sr & NV30_VP_SRC2_LOW_MASK) <<
+			  NV30_VP_INST_SRC2L_SHIFT;
+		break;
+	default:
+		assert(0);
+	}
+}
+
+static void
+emit_dst(struct nv20_vpc *vpc, uint32_t *hw, int slot, struct nv20_sreg dst)
+{
+	struct nv20_vertex_program *vp = vpc->vp;
+
+	switch (dst.type) {
+	case NV30SR_TEMP:
+		hw[0] |= (dst.index << NV30_VP_INST_DEST_TEMP_ID_SHIFT);
+		break;
+	case NV30SR_OUTPUT:
+		switch (dst.index) {
+		case NV30_VP_INST_DEST_COL0 : vp->or |= (1 << 0); break;
+		case NV30_VP_INST_DEST_COL1 : vp->or |= (1 << 1); break;
+		case NV30_VP_INST_DEST_BFC0 : vp->or |= (1 << 2); break;
+		case NV30_VP_INST_DEST_BFC1 : vp->or |= (1 << 3); break;
+		case NV30_VP_INST_DEST_FOGC : vp->or |= (1 << 4); break;
+		case NV30_VP_INST_DEST_PSZ  : vp->or |= (1 << 5); break;
+		case NV30_VP_INST_DEST_TC(0): vp->or |= (1 << 14); break;
+		case NV30_VP_INST_DEST_TC(1): vp->or |= (1 << 15); break;
+		case NV30_VP_INST_DEST_TC(2): vp->or |= (1 << 16); break;
+		case NV30_VP_INST_DEST_TC(3): vp->or |= (1 << 17); break;
+		case NV30_VP_INST_DEST_TC(4): vp->or |= (1 << 18); break;
+		case NV30_VP_INST_DEST_TC(5): vp->or |= (1 << 19); break;
+		case NV30_VP_INST_DEST_TC(6): vp->or |= (1 << 20); break;
+		case NV30_VP_INST_DEST_TC(7): vp->or |= (1 << 21); break;
+		default:
+			break;
+		}
+
+		hw[3] |= (dst.index << NV30_VP_INST_DEST_SHIFT);
+		hw[0] |= NV30_VP_INST_VEC_DEST_TEMP_MASK | (1<<20);
+
+		/*XXX: no way this is entirely correct, someone needs to
+		 *     figure out what exactly it is.
+		 */
+		hw[3] |= 0x800;
+		break;
+	default:
+		assert(0);
+	}
+}
+
+static void
+nv20_vp_arith(struct nv20_vpc *vpc, int slot, int op,
+	      struct nv20_sreg dst, int mask,
+	      struct nv20_sreg s0, struct nv20_sreg s1,
+	      struct nv20_sreg s2)
+{
+	struct nv20_vertex_program *vp = vpc->vp;
+	uint32_t *hw;
+
+	vp->insns = realloc(vp->insns, ++vp->nr_insns * sizeof(*vpc->vpi));
+	vpc->vpi = &vp->insns[vp->nr_insns - 1];
+	memset(vpc->vpi, 0, sizeof(*vpc->vpi));
+	vpc->vpi->const_index = -1;
+
+	hw = vpc->vpi->data;
+
+	hw[0] |= (NV30_VP_INST_COND_TR << NV30_VP_INST_COND_SHIFT);
+	hw[0] |= ((0 << NV30_VP_INST_COND_SWZ_X_SHIFT) |
+		  (1 << NV30_VP_INST_COND_SWZ_Y_SHIFT) |
+		  (2 << NV30_VP_INST_COND_SWZ_Z_SHIFT) |
+		  (3 << NV30_VP_INST_COND_SWZ_W_SHIFT));
+
+	hw[1] |= (op << NV30_VP_INST_VEC_OPCODE_SHIFT);
+//	hw[3] |= NV30_VP_INST_SCA_DEST_TEMP_MASK;
+//	hw[3] |= (mask << NV30_VP_INST_VEC_WRITEMASK_SHIFT);
+
+	if (dst.type == NV30SR_OUTPUT) {
+		if (slot)
+			hw[3] |= (mask << NV30_VP_INST_SDEST_WRITEMASK_SHIFT);
+		else
+			hw[3] |= (mask << NV30_VP_INST_VDEST_WRITEMASK_SHIFT);
+	} else {
+		if (slot)
+			hw[3] |= (mask << NV30_VP_INST_STEMP_WRITEMASK_SHIFT);
+		else
+			hw[3] |= (mask << NV30_VP_INST_VTEMP_WRITEMASK_SHIFT);
+	}
+
+	emit_dst(vpc, hw, slot, dst);
+	emit_src(vpc, hw, 0, s0);
+	emit_src(vpc, hw, 1, s1);
+	emit_src(vpc, hw, 2, s2);
+}
+
+static INLINE struct nv20_sreg
+tgsi_src(struct nv20_vpc *vpc, const struct tgsi_full_src_register *fsrc) {
+	struct nv20_sreg src;
+
+	switch (fsrc->SrcRegister.File) {
+	case TGSI_FILE_INPUT:
+		src = nv20_sr(NV30SR_INPUT, fsrc->SrcRegister.Index);
+		break;
+	case TGSI_FILE_CONSTANT:
+		src = constant(vpc, fsrc->SrcRegister.Index, 0, 0, 0, 0);
+		break;
+	case TGSI_FILE_IMMEDIATE:
+		src = vpc->imm[fsrc->SrcRegister.Index];
+		break;
+	case TGSI_FILE_TEMPORARY:
+		if (vpc->high_temp < fsrc->SrcRegister.Index)
+			vpc->high_temp = fsrc->SrcRegister.Index;
+		src = nv20_sr(NV30SR_TEMP, fsrc->SrcRegister.Index);
+		break;
+	default:
+		NOUVEAU_ERR("bad src file\n");
+		break;
+	}
+
+	src.abs = fsrc->SrcRegisterExtMod.Absolute;
+	src.negate = fsrc->SrcRegister.Negate;
+	src.swz[0] = fsrc->SrcRegister.SwizzleX;
+	src.swz[1] = fsrc->SrcRegister.SwizzleY;
+	src.swz[2] = fsrc->SrcRegister.SwizzleZ;
+	src.swz[3] = fsrc->SrcRegister.SwizzleW;
+	return src;
+}
+
+static INLINE struct nv20_sreg
+tgsi_dst(struct nv20_vpc *vpc, const struct tgsi_full_dst_register *fdst) {
+	struct nv20_sreg dst;
+
+	switch (fdst->DstRegister.File) {
+	case TGSI_FILE_OUTPUT:
+		dst = nv20_sr(NV30SR_OUTPUT,
+			      vpc->output_map[fdst->DstRegister.Index]);
+
+		break;
+	case TGSI_FILE_TEMPORARY:
+		dst = nv20_sr(NV30SR_TEMP, fdst->DstRegister.Index);
+		if (vpc->high_temp < dst.index)
+			vpc->high_temp = dst.index;
+		break;
+	default:
+		NOUVEAU_ERR("bad dst file\n");
+		break;
+	}
+
+	return dst;
+}
+
+static INLINE int
+tgsi_mask(uint tgsi)
+{
+	int mask = 0;
+
+	if (tgsi & TGSI_WRITEMASK_X) mask |= MASK_X;
+	if (tgsi & TGSI_WRITEMASK_Y) mask |= MASK_Y;
+	if (tgsi & TGSI_WRITEMASK_Z) mask |= MASK_Z;
+	if (tgsi & TGSI_WRITEMASK_W) mask |= MASK_W;
+	return mask;
+}
+
+static boolean
+nv20_vertprog_parse_instruction(struct nv20_vpc *vpc,
+				const struct tgsi_full_instruction *finst)
+{
+	struct nv20_sreg src[3], dst, tmp;
+	struct nv20_sreg none = nv20_sr(NV30SR_NONE, 0);
+	int mask;
+	int ai = -1, ci = -1;
+	int i;
+
+	if (finst->Instruction.Opcode == TGSI_OPCODE_END)
+		return TRUE;
+
+	vpc->temp_temp_count = 0;
+	for (i = 0; i < finst->Instruction.NumSrcRegs; i++) {
+		const struct tgsi_full_src_register *fsrc;
+
+		fsrc = &finst->FullSrcRegisters[i];
+		if (fsrc->SrcRegister.File == TGSI_FILE_TEMPORARY) {
+			src[i] = tgsi_src(vpc, fsrc);
+		}
+	}
+
+	for (i = 0; i < finst->Instruction.NumSrcRegs; i++) {
+		const struct tgsi_full_src_register *fsrc;
+
+		fsrc = &finst->FullSrcRegisters[i];
+		switch (fsrc->SrcRegister.File) {
+		case TGSI_FILE_INPUT:
+			if (ai == -1 || ai == fsrc->SrcRegister.Index) {
+				ai = fsrc->SrcRegister.Index;
+				src[i] = tgsi_src(vpc, fsrc);
+			} else {
+				src[i] = temp(vpc);
+				arith(vpc, 0, OP_MOV, src[i], MASK_ALL,
+				      tgsi_src(vpc, fsrc), none, none);
+			}
+			break;
+		/*XXX: index comparison is broken now that consts come from
+		 *     two different register files.
+		 */
+		case TGSI_FILE_CONSTANT:
+		case TGSI_FILE_IMMEDIATE:
+			if (ci == -1 || ci == fsrc->SrcRegister.Index) {
+				ci = fsrc->SrcRegister.Index;
+				src[i] = tgsi_src(vpc, fsrc);
+			} else {
+				src[i] = temp(vpc);
+				arith(vpc, 0, OP_MOV, src[i], MASK_ALL,
+				      tgsi_src(vpc, fsrc), none, none);
+			}
+			break;
+		case TGSI_FILE_TEMPORARY:
+			/* handled above */
+			break;
+		default:
+			NOUVEAU_ERR("bad src file\n");
+			return FALSE;
+		}
+	}
+
+	dst  = tgsi_dst(vpc, &finst->FullDstRegisters[0]);
+	mask = tgsi_mask(finst->FullDstRegisters[0].DstRegister.WriteMask);
+
+	switch (finst->Instruction.Opcode) {
+	case TGSI_OPCODE_ABS:
+		arith(vpc, 0, OP_MOV, dst, mask, abs(src[0]), none, none);
+		break;
+	case TGSI_OPCODE_ADD:
+		arith(vpc, 0, OP_ADD, dst, mask, src[0], none, src[1]);
+		break;
+	case TGSI_OPCODE_ARL:
+		arith(vpc, 0, OP_ARL, dst, mask, src[0], none, none);
+		break;
+	case TGSI_OPCODE_DP3:
+		arith(vpc, 0, OP_DP3, dst, mask, src[0], src[1], none);
+		break;
+	case TGSI_OPCODE_DP4:
+		arith(vpc, 0, OP_DP4, dst, mask, src[0], src[1], none);
+		break;
+	case TGSI_OPCODE_DPH:
+		arith(vpc, 0, OP_DPH, dst, mask, src[0], src[1], none);
+		break;
+	case TGSI_OPCODE_DST:
+		arith(vpc, 0, OP_DST, dst, mask, src[0], src[1], none);
+		break;
+	case TGSI_OPCODE_EX2:
+		arith(vpc, 1, OP_EX2, dst, mask, none, none, src[0]);
+		break;
+	case TGSI_OPCODE_EXP:
+		arith(vpc, 1, OP_EXP, dst, mask, none, none, src[0]);
+		break;
+	case TGSI_OPCODE_FLR:
+		arith(vpc, 0, OP_FLR, dst, mask, src[0], none, none);
+		break;
+	case TGSI_OPCODE_FRC:
+		arith(vpc, 0, OP_FRC, dst, mask, src[0], none, none);
+		break;
+	case TGSI_OPCODE_LG2:
+		arith(vpc, 1, OP_LG2, dst, mask, none, none, src[0]);
+		break;
+	case TGSI_OPCODE_LIT:
+		arith(vpc, 1, OP_LIT, dst, mask, none, none, src[0]);
+		break;
+	case TGSI_OPCODE_LOG:
+		arith(vpc, 1, OP_LOG, dst, mask, none, none, src[0]);
+		break;
+	case TGSI_OPCODE_MAD:
+		arith(vpc, 0, OP_MAD, dst, mask, src[0], src[1], src[2]);
+		break;
+	case TGSI_OPCODE_MAX:
+		arith(vpc, 0, OP_MAX, dst, mask, src[0], src[1], none);
+		break;
+	case TGSI_OPCODE_MIN:
+		arith(vpc, 0, OP_MIN, dst, mask, src[0], src[1], none);
+		break;
+	case TGSI_OPCODE_MOV:
+		arith(vpc, 0, OP_MOV, dst, mask, src[0], none, none);
+		break;
+	case TGSI_OPCODE_MUL:
+		arith(vpc, 0, OP_MUL, dst, mask, src[0], src[1], none);
+		break;
+	case TGSI_OPCODE_POW:
+		tmp = temp(vpc);
+		arith(vpc, 1, OP_LG2, tmp, MASK_X, none, none,
+		      swz(src[0], X, X, X, X));
+		arith(vpc, 0, OP_MUL, tmp, MASK_X, swz(tmp, X, X, X, X),
+		      swz(src[1], X, X, X, X), none);
+		arith(vpc, 1, OP_EX2, dst, mask, none, none,
+		      swz(tmp, X, X, X, X));
+		break;
+	case TGSI_OPCODE_RCP:
+		arith(vpc, 1, OP_RCP, dst, mask, none, none, src[0]);
+		break;
+	case TGSI_OPCODE_RET:
+		break;
+	case TGSI_OPCODE_RSQ:
+		arith(vpc, 1, OP_RSQ, dst, mask, none, none, src[0]);
+		break;
+	case TGSI_OPCODE_SGE:
+		arith(vpc, 0, OP_SGE, dst, mask, src[0], src[1], none);
+		break;
+	case TGSI_OPCODE_SGT:
+		arith(vpc, 0, OP_SGT, dst, mask, src[0], src[1], none);
+		break;
+	case TGSI_OPCODE_SLT:
+		arith(vpc, 0, OP_SLT, dst, mask, src[0], src[1], none);
+		break;
+	case TGSI_OPCODE_SUB:
+		arith(vpc, 0, OP_ADD, dst, mask, src[0], none, neg(src[1]));
+		break;
+	case TGSI_OPCODE_XPD:
+		tmp = temp(vpc);
+		arith(vpc, 0, OP_MUL, tmp, mask,
+		      swz(src[0], Z, X, Y, Y), swz(src[1], Y, Z, X, X), none);
+		arith(vpc, 0, OP_MAD, dst, (mask & ~MASK_W),
+		      swz(src[0], Y, Z, X, X), swz(src[1], Z, X, Y, Y),
+		      neg(tmp));
+		break;
+	default:
+		NOUVEAU_ERR("invalid opcode %d\n", finst->Instruction.Opcode);
+		return FALSE;
+	}
+
+	return TRUE;
+}
+
+static boolean
+nv20_vertprog_parse_decl_output(struct nv20_vpc *vpc,
+				const struct tgsi_full_declaration *fdec)
+{
+	int hw;
+
+	switch (fdec->Semantic.SemanticName) {
+	case TGSI_SEMANTIC_POSITION:
+		hw = NV30_VP_INST_DEST_POS;
+		break;
+	case TGSI_SEMANTIC_COLOR:
+		if (fdec->Semantic.SemanticIndex == 0) {
+			hw = NV30_VP_INST_DEST_COL0;
+		} else
+		if (fdec->Semantic.SemanticIndex == 1) {
+			hw = NV30_VP_INST_DEST_COL1;
+		} else {
+			NOUVEAU_ERR("bad colour semantic index\n");
+			return FALSE;
+		}
+		break;
+	case TGSI_SEMANTIC_BCOLOR:
+		if (fdec->Semantic.SemanticIndex == 0) {
+			hw = NV30_VP_INST_DEST_BFC0;
+		} else
+		if (fdec->Semantic.SemanticIndex == 1) {
+			hw = NV30_VP_INST_DEST_BFC1;
+		} else {
+			NOUVEAU_ERR("bad bcolour semantic index\n");
+			return FALSE;
+		}
+		break;
+	case TGSI_SEMANTIC_FOG:
+		hw = NV30_VP_INST_DEST_FOGC;
+		break;
+	case TGSI_SEMANTIC_PSIZE:
+		hw = NV30_VP_INST_DEST_PSZ;
+		break;
+	case TGSI_SEMANTIC_GENERIC:
+		if (fdec->Semantic.SemanticIndex <= 7) {
+			hw = NV30_VP_INST_DEST_TC(fdec->Semantic.SemanticIndex);
+		} else {
+			NOUVEAU_ERR("bad generic semantic index\n");
+			return FALSE;
+		}
+		break;
+	default:
+		NOUVEAU_ERR("bad output semantic\n");
+		return FALSE;
+	}
+
+	vpc->output_map[fdec->DeclarationRange.First] = hw;
+	return TRUE;
+}
+
+static boolean
+nv20_vertprog_prepare(struct nv20_vpc *vpc)
+{
+	struct tgsi_parse_context p;
+	int nr_imm = 0;
+
+	tgsi_parse_init(&p, vpc->vp->pipe.tokens);
+	while (!tgsi_parse_end_of_tokens(&p)) {
+		const union tgsi_full_token *tok = &p.FullToken;
+
+		tgsi_parse_token(&p);
+		switch(tok->Token.Type) {
+		case TGSI_TOKEN_TYPE_IMMEDIATE:
+			nr_imm++;
+			break;
+		default:
+			break;
+		}
+	}
+	tgsi_parse_free(&p);
+
+	if (nr_imm) {
+		vpc->imm = CALLOC(nr_imm, sizeof(struct nv20_sreg));
+		assert(vpc->imm);
+	}
+
+	return TRUE;
+}
+
+static void
+nv20_vertprog_translate(struct nv20_context *nv20,
+			struct nv20_vertex_program *vp)
+{
+	struct tgsi_parse_context parse;
+	struct nv20_vpc *vpc = NULL;
+
+	tgsi_dump(vp->pipe.tokens,0);
+
+	vpc = CALLOC(1, sizeof(struct nv20_vpc));
+	if (!vpc)
+		return;
+	vpc->vp = vp;
+	vpc->high_temp = -1;
+
+	if (!nv20_vertprog_prepare(vpc)) {
+		FREE(vpc);
+		return;
+	}
+
+	tgsi_parse_init(&parse, vp->pipe.tokens);
+
+	while (!tgsi_parse_end_of_tokens(&parse)) {
+		tgsi_parse_token(&parse);
+
+		switch (parse.FullToken.Token.Type) {
+		case TGSI_TOKEN_TYPE_DECLARATION:
+		{
+			const struct tgsi_full_declaration *fdec;
+			fdec = &parse.FullToken.FullDeclaration;
+			switch (fdec->Declaration.File) {
+			case TGSI_FILE_OUTPUT:
+				if (!nv20_vertprog_parse_decl_output(vpc, fdec))
+					goto out_err;
+				break;
+			default:
+				break;
+			}
+		}
+			break;
+		case TGSI_TOKEN_TYPE_IMMEDIATE:
+		{
+			const struct tgsi_full_immediate *imm;
+
+			imm = &parse.FullToken.FullImmediate;
+			assert(imm->Immediate.DataType == TGSI_IMM_FLOAT32);
+//			assert(imm->Immediate.Size == 4);
+			vpc->imm[vpc->nr_imm++] =
+				constant(vpc, -1,
+					 imm->u.ImmediateFloat32[0].Float,
+					 imm->u.ImmediateFloat32[1].Float,
+					 imm->u.ImmediateFloat32[2].Float,
+					 imm->u.ImmediateFloat32[3].Float);
+		}
+			break;
+		case TGSI_TOKEN_TYPE_INSTRUCTION:
+		{
+			const struct tgsi_full_instruction *finst;
+			finst = &parse.FullToken.FullInstruction;
+			if (!nv20_vertprog_parse_instruction(vpc, finst))
+				goto out_err;
+		}
+			break;
+		default:
+			break;
+		}
+	}
+
+	vp->insns[vp->nr_insns - 1].data[3] |= NV30_VP_INST_LAST;
+	vp->translated = TRUE;
+out_err:
+	tgsi_parse_free(&parse);
+	FREE(vpc);
+}
+
+static boolean
+nv20_vertprog_validate(struct nv20_context *nv20)
+{ 
+	struct nouveau_winsys *nvws = nv20->nvws;
+	struct pipe_winsys *ws = nv20->pipe.winsys;
+	struct nouveau_grobj *rankine = nv20->screen->rankine;
+	struct nv20_vertex_program *vp;
+	struct pipe_buffer *constbuf;
+	boolean upload_code = FALSE, upload_data = FALSE;
+	int i;
+
+	vp = nv20->vertprog;
+	constbuf = nv20->constbuf[PIPE_SHADER_VERTEX];
+
+	/* Translate TGSI shader into hw bytecode */
+	if (!vp->translated) {
+		nv20_vertprog_translate(nv20, vp);
+		if (!vp->translated)
+			return FALSE;
+	}
+
+	/* Allocate hw vtxprog exec slots */
+	if (!vp->exec) {
+		struct nouveau_resource *heap = nv20->screen->vp_exec_heap;
+		struct nouveau_stateobj *so;
+		uint vplen = vp->nr_insns;
+
+		if (nvws->res_alloc(heap, vplen, vp, &vp->exec)) {
+			while (heap->next && heap->size < vplen) {
+				struct nv20_vertex_program *evict;
+				
+				evict = heap->next->priv;
+				nvws->res_free(&evict->exec);
+			}
+
+			if (nvws->res_alloc(heap, vplen, vp, &vp->exec))
+				assert(0);
+		}
+
+		so = so_new(2, 0);
+		so_method(so, rankine, NV34TCL_VP_START_FROM_ID, 1);
+		so_data  (so, vp->exec->start);
+		so_ref(so, &vp->so);
+
+		upload_code = TRUE;
+	}
+
+	/* Allocate hw vtxprog const slots */
+	if (vp->nr_consts && !vp->data) {
+		struct nouveau_resource *heap = nv20->screen->vp_data_heap;
+
+		if (nvws->res_alloc(heap, vp->nr_consts, vp, &vp->data)) {
+			while (heap->next && heap->size < vp->nr_consts) {
+				struct nv20_vertex_program *evict;
+				
+				evict = heap->next->priv;
+				nvws->res_free(&evict->data);
+			}
+
+			if (nvws->res_alloc(heap, vp->nr_consts, vp, &vp->data))
+				assert(0);
+		}
+
+		/*XXX: handle this some day */
+		assert(vp->data->start >= vp->data_start_min);
+
+		upload_data = TRUE;
+		if (vp->data_start != vp->data->start)
+			upload_code = TRUE;
+	}
+
+	/* If exec or data segments moved we need to patch the program to
+	 * fixup offsets and register IDs.
+	 */
+	if (vp->exec_start != vp->exec->start) {
+		for (i = 0; i < vp->nr_insns; i++) {
+			struct nv20_vertex_program_exec *vpi = &vp->insns[i];
+
+			if (vpi->has_branch_offset) {
+				assert(0);
+			}
+		}
+
+		vp->exec_start = vp->exec->start;
+	}
+
+	if (vp->nr_consts && vp->data_start != vp->data->start) {
+		for (i = 0; i < vp->nr_insns; i++) {
+			struct nv20_vertex_program_exec *vpi = &vp->insns[i];
+
+			if (vpi->const_index >= 0) {
+				vpi->data[1] &= ~NV30_VP_INST_CONST_SRC_MASK;
+				vpi->data[1] |=
+					(vpi->const_index + vp->data->start) <<
+					NV30_VP_INST_CONST_SRC_SHIFT;
+
+			}
+		}
+
+		vp->data_start = vp->data->start;
+	}
+
+	/* Update + Upload constant values */
+	if (vp->nr_consts) {
+		float *map = NULL;
+
+		if (constbuf) {
+			map = ws->buffer_map(ws, constbuf,
+					     PIPE_BUFFER_USAGE_CPU_READ);
+		}
+
+		for (i = 0; i < vp->nr_consts; i++) {
+			struct nv20_vertex_program_data *vpd = &vp->consts[i];
+
+			if (vpd->index >= 0) {
+				if (!upload_data &&
+				    !memcmp(vpd->value, &map[vpd->index * 4],
+					    4 * sizeof(float)))
+					continue;
+				memcpy(vpd->value, &map[vpd->index * 4],
+				       4 * sizeof(float));
+			}
+
+			BEGIN_RING(rankine, NV34TCL_VP_UPLOAD_CONST_ID, 5);
+			OUT_RING  (i + vp->data->start);
+			OUT_RINGp ((uint32_t *)vpd->value, 4);
+		}
+
+		if (constbuf) {
+			ws->buffer_unmap(ws, constbuf);
+		}
+	}
+
+	/* Upload vtxprog */
+	if (upload_code) {
+#if 0
+		for (i = 0; i < vp->nr_insns; i++) {
+			NOUVEAU_MSG("VP inst %d: 0x%08x 0x%08x 0x%08x 0x%08x\n",
+				i, vp->insns[i].data[0], vp->insns[i].data[1],
+				vp->insns[i].data[2], vp->insns[i].data[3]);
+		}
+#endif
+		BEGIN_RING(rankine, NV34TCL_VP_UPLOAD_FROM_ID, 1);
+		OUT_RING  (vp->exec->start);
+		for (i = 0; i < vp->nr_insns; i++) {
+			BEGIN_RING(rankine, NV34TCL_VP_UPLOAD_INST(0), 4);
+			OUT_RINGp (vp->insns[i].data, 4);
+		}
+	}
+
+	if (vp->so != nv20->state.hw[NV30_STATE_VERTPROG]) {
+		so_ref(vp->so, &nv20->state.hw[NV30_STATE_VERTPROG]);
+		return TRUE;
+	}
+
+	return FALSE;
+}
+
+void
+nv20_vertprog_destroy(struct nv20_context *nv20, struct nv20_vertex_program *vp)
+{
+	struct nouveau_winsys *nvws = nv20->screen->nvws;
+
+	vp->translated = FALSE;
+
+	if (vp->nr_insns) {
+		FREE(vp->insns);
+		vp->insns = NULL;
+		vp->nr_insns = 0;
+	}
+
+	if (vp->nr_consts) {
+		FREE(vp->consts);
+		vp->consts = NULL;
+		vp->nr_consts = 0;
+	}
+
+	nvws->res_free(&vp->exec);
+	vp->exec_start = 0;
+	nvws->res_free(&vp->data);
+	vp->data_start = 0;
+	vp->data_start_min = 0;
+
+	vp->ir = vp->or = 0;
+	so_ref(NULL, &vp->so);
+}
+
+struct nv20_state_entry nv20_state_vertprog = {
+	.validate = nv20_vertprog_validate,
+	.dirty = {
+		.pipe = NV30_NEW_VERTPROG /*| NV30_NEW_UCP*/,
+		.hw = NV30_STATE_VERTPROG,
+	}
+};