Merge remote branch 'origin/master' into lp-binning

Conflicts:
	src/gallium/auxiliary/util/u_surface.c
	src/gallium/drivers/llvmpipe/Makefile
	src/gallium/drivers/llvmpipe/SConscript
	src/gallium/drivers/llvmpipe/lp_bld_arit.c
	src/gallium/drivers/llvmpipe/lp_bld_flow.c
	src/gallium/drivers/llvmpipe/lp_bld_interp.c
	src/gallium/drivers/llvmpipe/lp_clear.c
	src/gallium/drivers/llvmpipe/lp_context.c
	src/gallium/drivers/llvmpipe/lp_context.h
	src/gallium/drivers/llvmpipe/lp_draw_arrays.c
	src/gallium/drivers/llvmpipe/lp_jit.c
	src/gallium/drivers/llvmpipe/lp_jit.h
	src/gallium/drivers/llvmpipe/lp_prim_vbuf.c
	src/gallium/drivers/llvmpipe/lp_setup.c
	src/gallium/drivers/llvmpipe/lp_setup_point.c
	src/gallium/drivers/llvmpipe/lp_state.h
	src/gallium/drivers/llvmpipe/lp_state_blend.c
	src/gallium/drivers/llvmpipe/lp_state_derived.c
	src/gallium/drivers/llvmpipe/lp_state_fs.c
	src/gallium/drivers/llvmpipe/lp_state_sampler.c
	src/gallium/drivers/llvmpipe/lp_state_surface.c
	src/gallium/drivers/llvmpipe/lp_tex_cache.c
	src/gallium/drivers/llvmpipe/lp_tex_cache.h
	src/gallium/drivers/llvmpipe/lp_tex_sample.h
	src/gallium/drivers/llvmpipe/lp_tile_cache.c

13 files changed, 293 insertions, 215 deletions

diff --git a/src/gallium/drivers/nv04/nv04_context.c b/src/gallium/drivers/nv04/nv04_context.c
index 17166c9f51..edd96859cf 100644
--- a/src/gallium/drivers/nv04/nv04_context.c
+++ b/src/gallium/drivers/nv04/nv04_context.c
@@ -10,10 +10,14 @@ nv04_flush(struct pipe_context *pipe, unsigned flags,
 	   struct pipe_fence_handle **fence)
 {
 	struct nv04_context *nv04 = nv04_context(pipe);
+	struct nv04_screen *screen = nv04->screen;
+	struct nouveau_channel *chan = screen->base.channel;
 
 	draw_flush(nv04->draw);
 
-	FIRE_RING(fence);
+	FIRE_RING(chan);
+	if (fence)
+		*fence = NULL;
 }
 
 static void
@@ -27,67 +31,42 @@ nv04_destroy(struct pipe_context *pipe)
 	FREE(nv04);
 }
 
-static void
-nv04_set_edgeflags(struct pipe_context *pipe, const unsigned *bitfield)
-{
-}
-
 static boolean
 nv04_init_hwctx(struct nv04_context *nv04)
 {
+	struct nv04_screen *screen = nv04->screen;
+	struct nouveau_channel *chan = screen->base.channel;
+	struct nouveau_grobj *fahrenheit = screen->fahrenheit;
+
 	// requires a valid handle
-//	BEGIN_RING(fahrenheit, NV04_DX5_TEXTURED_TRIANGLE_NOTIFY, 1);
+//	BEGIN_RING(chan, fahrenheit, NV04_TEXTURED_TRIANGLE_NOTIFY, 1);
 //	OUT_RING(0);
-	BEGIN_RING(fahrenheit, NV04_DX5_TEXTURED_TRIANGLE_NOP, 1);
-	OUT_RING(0);
+	BEGIN_RING(chan, fahrenheit, NV04_TEXTURED_TRIANGLE_NOP, 1);
+	OUT_RING(chan, 0);
 
-	BEGIN_RING(fahrenheit, NV04_DX5_TEXTURED_TRIANGLE_CONTROL, 1);
-	OUT_RING(0x40182800);
+	BEGIN_RING(chan, fahrenheit, NV04_TEXTURED_TRIANGLE_CONTROL, 1);
+	OUT_RING(chan, 0x40182800);
 //	OUT_RING(1<<20/*no cull*/);
-	BEGIN_RING(fahrenheit, NV04_DX5_TEXTURED_TRIANGLE_BLEND, 1);
+	BEGIN_RING(chan, fahrenheit, NV04_TEXTURED_TRIANGLE_BLEND, 1);
 //	OUT_RING(0x24|(1<<6)|(1<<8));
-	OUT_RING(0x120001a4);
-	BEGIN_RING(fahrenheit, NV04_DX5_TEXTURED_TRIANGLE_FORMAT, 1);
-	OUT_RING(0x332213a1);
-	BEGIN_RING(fahrenheit, NV04_DX5_TEXTURED_TRIANGLE_FILTER, 1);
-	OUT_RING(0x11001010);
-	BEGIN_RING(fahrenheit, NV04_DX5_TEXTURED_TRIANGLE_COLORKEY, 1);
-	OUT_RING(0x0);
-//	BEGIN_RING(fahrenheit, NV04_DX5_TEXTURED_TRIANGLE_OFFSET, 1);
+	OUT_RING(chan, 0x120001a4);
+	BEGIN_RING(chan, fahrenheit, NV04_TEXTURED_TRIANGLE_FORMAT, 1);
+	OUT_RING(chan, 0x332213a1);
+	BEGIN_RING(chan, fahrenheit, NV04_TEXTURED_TRIANGLE_FILTER, 1);
+	OUT_RING(chan, 0x11001010);
+	BEGIN_RING(chan, fahrenheit, NV04_TEXTURED_TRIANGLE_COLORKEY, 1);
+	OUT_RING(chan, 0x0);
+//	BEGIN_RING(chan, fahrenheit, NV04_TEXTURED_TRIANGLE_OFFSET, 1);
 //	OUT_RING(SCREEN_OFFSET);
-	BEGIN_RING(fahrenheit, NV04_DX5_TEXTURED_TRIANGLE_FOGCOLOR, 1);
-	OUT_RING(0xff000000);
+	BEGIN_RING(chan, fahrenheit, NV04_TEXTURED_TRIANGLE_FOGCOLOR, 1);
+	OUT_RING(chan, 0xff000000);
 
 
 
-	FIRE_RING (NULL);
+	FIRE_RING (chan);
 	return TRUE;
 }
 
-static unsigned int
-nv04_is_texture_referenced( struct pipe_context *pipe,
-			    struct pipe_texture *texture,
-			    unsigned face, unsigned level)
-{
-   /**
-    * FIXME: Optimize.
-    */
-
-   return PIPE_REFERENCED_FOR_READ | PIPE_REFERENCED_FOR_WRITE;
-}
-
-static unsigned int
-nv04_is_buffer_referenced( struct pipe_context *pipe,
-			   struct pipe_buffer *buf)
-{
-   /**
-    * FIXME: Optimize.
-    */
-
-   return PIPE_REFERENCED_FOR_READ | PIPE_REFERENCED_FOR_WRITE;
-}
-
-
 struct pipe_context *
 nv04_create(struct pipe_screen *pscreen, unsigned pctx_id)
 {
@@ -107,14 +86,13 @@ nv04_create(struct pipe_screen *pscreen, unsigned pctx_id)
 	nv04->pipe.winsys = ws;
 	nv04->pipe.screen = pscreen;
 	nv04->pipe.destroy = nv04_destroy;
-	nv04->pipe.set_edgeflags = nv04_set_edgeflags;
 	nv04->pipe.draw_arrays = nv04_draw_arrays;
 	nv04->pipe.draw_elements = nv04_draw_elements;
 	nv04->pipe.clear = nv04_clear;
 	nv04->pipe.flush = nv04_flush;
 
-	nv04->pipe.is_texture_referenced = nv04_is_texture_referenced;
-	nv04->pipe.is_buffer_referenced = nv04_is_buffer_referenced;
+	nv04->pipe.is_texture_referenced = nouveau_is_texture_referenced;
+	nv04->pipe.is_buffer_referenced = nouveau_is_buffer_referenced;
 
 	nv04_init_surface_functions(nv04);
 	nv04_init_state_functions(nv04);
diff --git a/src/gallium/drivers/nv04/nv04_context.h b/src/gallium/drivers/nv04/nv04_context.h
index 2842b2c90d..fe3b527423 100644
--- a/src/gallium/drivers/nv04/nv04_context.h
+++ b/src/gallium/drivers/nv04/nv04_context.h
@@ -13,10 +13,7 @@
 
 #include "nouveau/nouveau_winsys.h"
 #include "nouveau/nouveau_gldefs.h"
-
-#define NOUVEAU_PUSH_CONTEXT(ctx)                                              \
-	struct nv04_screen *ctx = nv04->screen
-#include "nouveau/nouveau_push.h"
+#include "nouveau/nouveau_context.h"
 
 #include "nv04_state.h"
 
@@ -140,9 +137,9 @@ extern void nv04_emit_hw_state(struct nv04_context *nv04);
 extern void nv04_state_tex_update(struct nv04_context *nv04);
 
 /* nv04_vbo.c */
-extern boolean nv04_draw_arrays(struct pipe_context *, unsigned mode,
+extern void nv04_draw_arrays(struct pipe_context *, unsigned mode,
 				unsigned start, unsigned count);
-extern boolean nv04_draw_elements( struct pipe_context *pipe,
+extern void nv04_draw_elements( struct pipe_context *pipe,
                     struct pipe_buffer *indexBuffer,
                     unsigned indexSize,
                     unsigned prim, unsigned start, unsigned count);
diff --git a/src/gallium/drivers/nv04/nv04_fragtex.c b/src/gallium/drivers/nv04/nv04_fragtex.c
index 21f990fd53..c152b52119 100644
--- a/src/gallium/drivers/nv04/nv04_fragtex.c
+++ b/src/gallium/drivers/nv04/nv04_fragtex.c
@@ -4,7 +4,7 @@
 #define _(m,tf)                                                                \
 {                                                                              \
   PIPE_FORMAT_##m,                                                             \
-  NV04_DX5_TEXTURED_TRIANGLE_FORMAT_COLOR_##tf,                                               \
+  NV04_TEXTURED_TRIANGLE_FORMAT_COLOR_##tf,                                               \
 }
 
 struct nv04_texture_format {
@@ -53,14 +53,14 @@ nv04_fragtex_build(struct nv04_context *nv04, int unit)
 		return;
 	}
 
-	nv04->fragtex.format = NV04_DX5_TEXTURED_TRIANGLE_FORMAT_ORIGIN_ZOH_CORNER 
-		| NV04_DX5_TEXTURED_TRIANGLE_FORMAT_ORIGIN_FOH_CORNER
+	nv04->fragtex.format = NV04_TEXTURED_TRIANGLE_FORMAT_ORIGIN_ZOH_CORNER
+		| NV04_TEXTURED_TRIANGLE_FORMAT_ORIGIN_FOH_CORNER
 		| nv04_fragtex_format(pt->format)
-		| ( (pt->last_level + 1) << NV04_DX5_TEXTURED_TRIANGLE_FORMAT_MIPMAP_LEVELS_SHIFT )
-		| ( log2i(pt->width[0]) << NV04_DX5_TEXTURED_TRIANGLE_FORMAT_BASE_SIZE_U_SHIFT )
-		| ( log2i(pt->height[0]) << NV04_DX5_TEXTURED_TRIANGLE_FORMAT_BASE_SIZE_V_SHIFT )
-		| NV04_DX5_TEXTURED_TRIANGLE_FORMAT_ADDRESSU_CLAMP_TO_EDGE
-		| NV04_DX5_TEXTURED_TRIANGLE_FORMAT_ADDRESSV_CLAMP_TO_EDGE
+		| ( (pt->last_level + 1) << NV04_TEXTURED_TRIANGLE_FORMAT_MIPMAP_LEVELS_SHIFT )
+		| ( log2i(pt->width0) << NV04_TEXTURED_TRIANGLE_FORMAT_BASE_SIZE_U_SHIFT )
+		| ( log2i(pt->height0) << NV04_TEXTURED_TRIANGLE_FORMAT_BASE_SIZE_V_SHIFT )
+		| NV04_TEXTURED_TRIANGLE_FORMAT_ADDRESSU_CLAMP_TO_EDGE
+		| NV04_TEXTURED_TRIANGLE_FORMAT_ADDRESSV_CLAMP_TO_EDGE
 		;
 }
 
diff --git a/src/gallium/drivers/nv04/nv04_miptree.c b/src/gallium/drivers/nv04/nv04_miptree.c
index 93f752faec..e0a6948aeb 100644
--- a/src/gallium/drivers/nv04/nv04_miptree.c
+++ b/src/gallium/drivers/nv04/nv04_miptree.c
@@ -1,6 +1,7 @@
 #include "pipe/p_state.h"
 #include "pipe/p_defines.h"
 #include "pipe/p_inlines.h"
+#include "util/u_math.h"
 
 #include "nv04_context.h"
 #include "nv04_screen.h"
@@ -9,31 +10,22 @@ static void
 nv04_miptree_layout(struct nv04_miptree *nv04mt)
 {
 	struct pipe_texture *pt = &nv04mt->base;
-	uint width = pt->width[0], height = pt->height[0];
 	uint offset = 0;
 	int nr_faces, l;
 
 	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);
-		
-		nv04mt->level[l].pitch = pt->width[0];
+		nv04mt->level[l].pitch = pt->width0;
 		nv04mt->level[l].pitch = (nv04mt->level[l].pitch + 63) & ~63;
-
-		width  = MAX2(1, width  >> 1);
-		height = MAX2(1, height >> 1);
 	}
 
 	for (l = 0; l <= pt->last_level; l++) {
-
 		nv04mt->level[l].image_offset = 
 			CALLOC(nr_faces, sizeof(unsigned));
-		offset += nv04mt->level[l].pitch * pt->height[l];
+		/* XXX guess was obviously missing */
+		nv04mt->level[l].image_offset[0] = offset;
+		offset += nv04mt->level[l].pitch * u_minify(pt->height0, l);
 	}
 
 	nv04mt->total_size = offset;
@@ -63,7 +55,7 @@ nv04_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *pt)
 		FREE(mt);
 		return NULL;
 	}
-	
+	mt->bo = nouveau_bo(mt->buffer);
 	return &mt->base;
 }
 
@@ -75,7 +67,7 @@ nv04_miptree_blanket(struct pipe_screen *pscreen, const struct pipe_texture *pt,
 
 	/* Only supports 2D, non-mipmapped textures for the moment */
 	if (pt->target != PIPE_TEXTURE_2D || pt->last_level != 0 ||
-	    pt->depth[0] != 1)
+	    pt->depth0 != 1)
 		return NULL;
 
 	mt = CALLOC_STRUCT(nv04_miptree);
@@ -89,6 +81,7 @@ nv04_miptree_blanket(struct pipe_screen *pscreen, const struct pipe_texture *pt,
 	mt->level[0].image_offset = CALLOC(1, sizeof(unsigned));
 
 	pipe_buffer_reference(&mt->buffer, pb);
+	mt->bo = nouveau_bo(mt->buffer);
 	return &mt->base;
 }
 
@@ -120,8 +113,8 @@ nv04_miptree_surface_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 		return NULL;
 	pipe_texture_reference(&ns->base.texture, pt);
 	ns->base.format = pt->format;
-	ns->base.width = pt->width[level];
-	ns->base.height = pt->height[level];
+	ns->base.width = u_minify(pt->width0, level);
+	ns->base.height = u_minify(pt->height0, level);
 	ns->base.usage = flags;
 	pipe_reference_init(&ns->base.reference, 1);
 	ns->base.face = face;
@@ -129,7 +122,7 @@ nv04_miptree_surface_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 	ns->base.zslice = zslice;
 	ns->pitch = nv04mt->level[level].pitch;
 
-	ns->base.offset = nv04mt->level[level].image_offset;
+	ns->base.offset = nv04mt->level[level].image_offset[0];
 
 	return &ns->base;
 }
diff --git a/src/gallium/drivers/nv04/nv04_prim_vbuf.c b/src/gallium/drivers/nv04/nv04_prim_vbuf.c
index f6458232ae..0b795ea243 100644
--- a/src/gallium/drivers/nv04/nv04_prim_vbuf.c
+++ b/src/gallium/drivers/nv04/nv04_prim_vbuf.c
@@ -93,33 +93,45 @@ nv04_vbuf_render_set_primitive( struct vbuf_render *render,
 
 static INLINE void nv04_2triangles(struct nv04_context* nv04, unsigned char* buffer, ushort v0, ushort v1, ushort v2, ushort v3, ushort v4, ushort v5)
 {
-	BEGIN_RING(fahrenheit,NV04_DX5_TEXTURED_TRIANGLE_TLVERTEX_SX(0xA),49);
-	OUT_RINGp(buffer + VERTEX_SIZE * v0,8);
-	OUT_RINGp(buffer + VERTEX_SIZE * v1,8);
-	OUT_RINGp(buffer + VERTEX_SIZE * v2,8);
-	OUT_RINGp(buffer + VERTEX_SIZE * v3,8);
-	OUT_RINGp(buffer + VERTEX_SIZE * v4,8);
-	OUT_RINGp(buffer + VERTEX_SIZE * v5,8);
-	OUT_RING(0xFEDCBA);
+	struct nv04_screen *screen = nv04->screen;
+	struct nouveau_channel *chan = screen->base.channel;
+	struct nouveau_grobj *fahrenheit = screen->fahrenheit;
+
+	BEGIN_RING(chan, fahrenheit, NV04_TEXTURED_TRIANGLE_TLVERTEX_SX(0xA), 49);
+	OUT_RINGp(chan, buffer + VERTEX_SIZE * v0,8);
+	OUT_RINGp(chan, buffer + VERTEX_SIZE * v1,8);
+	OUT_RINGp(chan, buffer + VERTEX_SIZE * v2,8);
+	OUT_RINGp(chan, buffer + VERTEX_SIZE * v3,8);
+	OUT_RINGp(chan, buffer + VERTEX_SIZE * v4,8);
+	OUT_RINGp(chan, buffer + VERTEX_SIZE * v5,8);
+	OUT_RING(chan, 0xFEDCBA);
 }
 
 static INLINE void nv04_1triangle(struct nv04_context* nv04, unsigned char* buffer, ushort v0, ushort v1, ushort v2)
 {
-	BEGIN_RING(fahrenheit,NV04_DX5_TEXTURED_TRIANGLE_TLVERTEX_SX(0xD),25);
-	OUT_RINGp(buffer + VERTEX_SIZE * v0,8);
-	OUT_RINGp(buffer + VERTEX_SIZE * v1,8);
-	OUT_RINGp(buffer + VERTEX_SIZE * v2,8);
-	OUT_RING(0xFED);
+	struct nv04_screen *screen = nv04->screen;
+	struct nouveau_channel *chan = screen->base.channel;
+	struct nouveau_grobj *fahrenheit = screen->fahrenheit;
+
+	BEGIN_RING(chan, fahrenheit, NV04_TEXTURED_TRIANGLE_TLVERTEX_SX(0xD), 25);
+	OUT_RINGp(chan, buffer + VERTEX_SIZE * v0,8);
+	OUT_RINGp(chan, buffer + VERTEX_SIZE * v1,8);
+	OUT_RINGp(chan, buffer + VERTEX_SIZE * v2,8);
+	OUT_RING(chan, 0xFED);
 }
 
 static INLINE void nv04_1quad(struct nv04_context* nv04, unsigned char* buffer, ushort v0, ushort v1, ushort v2, ushort v3)
 {
-	BEGIN_RING(fahrenheit,NV04_DX5_TEXTURED_TRIANGLE_TLVERTEX_SX(0xC),33);
-	OUT_RINGp(buffer + VERTEX_SIZE * v0,8);
-	OUT_RINGp(buffer + VERTEX_SIZE * v1,8);
-	OUT_RINGp(buffer + VERTEX_SIZE * v2,8);
-	OUT_RINGp(buffer + VERTEX_SIZE * v3,8);
-	OUT_RING(0xFECEDC);
+	struct nv04_screen *screen = nv04->screen;
+	struct nouveau_channel *chan = screen->base.channel;
+	struct nouveau_grobj *fahrenheit = screen->fahrenheit;
+
+	BEGIN_RING(chan, fahrenheit, NV04_TEXTURED_TRIANGLE_TLVERTEX_SX(0xC), 33);
+	OUT_RINGp(chan, buffer + VERTEX_SIZE * v0,8);
+	OUT_RINGp(chan, buffer + VERTEX_SIZE * v1,8);
+	OUT_RINGp(chan, buffer + VERTEX_SIZE * v2,8);
+	OUT_RINGp(chan, buffer + VERTEX_SIZE * v3,8);
+	OUT_RING(chan, 0xFECEDC);
 }
 
 static void nv04_vbuf_render_triangles_elts(struct nv04_vbuf_render * render, const ushort * indices, uint nr_indices)
@@ -156,7 +168,10 @@ static void nv04_vbuf_render_tri_strip_elts(struct nv04_vbuf_render* render, con
 {
 	const uint32_t striptbl[]={0x321210,0x543432,0x765654,0x987876,0xBA9A98,0xDCBCBA,0xFEDEDC};
 	unsigned char* buffer = render->buffer;
-	struct nv04_context* nv04 = render->nv04;
+	struct nv04_context *nv04 = render->nv04;
+	struct nv04_screen *screen = nv04->screen;
+	struct nouveau_channel *chan = screen->base.channel;
+	struct nouveau_grobj *fahrenheit = screen->fahrenheit;
 	int i,j;
 
 	for(i = 0; i<nr_indices; i+=14) 
@@ -166,15 +181,15 @@ static void nv04_vbuf_render_tri_strip_elts(struct nv04_vbuf_render* render, con
 		if (numvert<3)
 			break;
 
-		BEGIN_RING( fahrenheit, NV04_DX5_TEXTURED_TRIANGLE_TLVERTEX_SX(0x0), numvert*8 );
+		BEGIN_RING(chan, fahrenheit, NV04_TEXTURED_TRIANGLE_TLVERTEX_SX(0x0), numvert*8);
 		for(j = 0; j<numvert; j++)
-			OUT_RINGp( buffer + VERTEX_SIZE * indices [i+j], 8 );
+			OUT_RINGp(chan, buffer + VERTEX_SIZE * indices [i+j], 8 );
 
-		BEGIN_RING_NI( fahrenheit, NV04_DX5_TEXTURED_TRIANGLE_TLVERTEX_DRAWPRIMITIVE(0), (numtri+1)/2 );
+		BEGIN_RING_NI(chan, fahrenheit, NV04_TEXTURED_TRIANGLE_DRAWPRIMITIVE(0), (numtri+1)/2 );
 		for(j = 0; j<numtri/2; j++ )
-			OUT_RING(striptbl[j]);
+			OUT_RING(chan, striptbl[j]);
 		if (numtri%2)
-			OUT_RING(striptbl[numtri/2]&0xFFF);
+			OUT_RING(chan, striptbl[numtri/2]&0xFFF);
 	}
 }
 
@@ -182,11 +197,14 @@ static void nv04_vbuf_render_tri_fan_elts(struct nv04_vbuf_render* render, const
 {
 	const uint32_t fantbl[]={0x320210,0x540430,0x760650,0x980870,0xBA0A90,0xDC0CB0,0xFE0ED0};
 	unsigned char* buffer = render->buffer;
-	struct nv04_context* nv04 = render->nv04;
+	struct nv04_context *nv04 = render->nv04;
+	struct nv04_screen *screen = nv04->screen;
+	struct nouveau_channel *chan = screen->base.channel;
+	struct nouveau_grobj *fahrenheit = screen->fahrenheit;
 	int i,j;
 
-	BEGIN_RING(fahrenheit, NV04_DX5_TEXTURED_TRIANGLE_TLVERTEX_SX(0x0), 8);
-	OUT_RINGp(buffer + VERTEX_SIZE * indices[0], 8);
+	BEGIN_RING(chan, fahrenheit, NV04_TEXTURED_TRIANGLE_TLVERTEX_SX(0x0), 8);
+	OUT_RINGp(chan, buffer + VERTEX_SIZE * indices[0], 8);
 
 	for(i = 1; i<nr_indices; i+=14)
 	{
@@ -195,16 +213,16 @@ static void nv04_vbuf_render_tri_fan_elts(struct nv04_vbuf_render* render, const
 		if (numvert < 3)
 			break;
 
-		BEGIN_RING(fahrenheit, NV04_DX5_TEXTURED_TRIANGLE_TLVERTEX_SX(0x1), numvert*8);
+		BEGIN_RING(chan, fahrenheit, NV04_TEXTURED_TRIANGLE_TLVERTEX_SX(0x1), numvert*8);
 
 		for(j=0;j<numvert;j++)
-			OUT_RINGp( buffer + VERTEX_SIZE * indices[ i+j ], 8 );
+			OUT_RINGp(chan, buffer + VERTEX_SIZE * indices[ i+j ], 8 );
 
-		BEGIN_RING_NI(fahrenheit, NV04_DX5_TEXTURED_TRIANGLE_TLVERTEX_DRAWPRIMITIVE(0), (numtri+1)/2);
+		BEGIN_RING_NI(chan, fahrenheit, NV04_TEXTURED_TRIANGLE_DRAWPRIMITIVE(0), (numtri+1)/2);
 		for(j = 0; j<numtri/2; j++)
-			OUT_RING(fantbl[j]);
+			OUT_RING(chan, fantbl[j]);
 		if (numtri%2)
-			OUT_RING(fantbl[numtri/2]&0xFFF);
+			OUT_RING(chan, fantbl[numtri/2]&0xFFF);
 	}
 }
 
diff --git a/src/gallium/drivers/nv04/nv04_screen.c b/src/gallium/drivers/nv04/nv04_screen.c
index 170ce3eb7e..7c5b6e8229 100644
--- a/src/gallium/drivers/nv04/nv04_screen.c
+++ b/src/gallium/drivers/nv04/nv04_screen.c
@@ -119,6 +119,8 @@ nv04_screen_destroy(struct pipe_screen *pscreen)
 	nouveau_grobj_free(&screen->fahrenheit);
 	nv04_surface_2d_takedown(&screen->eng2d);
 
+	nouveau_screen_fini(&screen->base);
+
 	FREE(pscreen);
 }
 
@@ -163,10 +165,10 @@ nv04_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
 		fahrenheit_class = 0;
 		sub3d_class = 0;
 	} else if (dev->chipset >= 0x10) {
-		fahrenheit_class = NV10_DX5_TEXTURED_TRIANGLE;
+		fahrenheit_class = NV10_TEXTURED_TRIANGLE;
 		sub3d_class = NV10_CONTEXT_SURFACES_3D;
 	} else {
-		fahrenheit_class=NV04_DX5_TEXTURED_TRIANGLE;
+		fahrenheit_class=NV04_TEXTURED_TRIANGLE;
 		sub3d_class = NV04_CONTEXT_SURFACES_3D;
 	}
 
diff --git a/src/gallium/drivers/nv04/nv04_state.c b/src/gallium/drivers/nv04/nv04_state.c
index d356ebd8b3..e3dc4c5bf4 100644
--- a/src/gallium/drivers/nv04/nv04_state.c
+++ b/src/gallium/drivers/nv04/nv04_state.c
@@ -50,28 +50,28 @@ wrap_mode(unsigned wrap) {
 
 	switch (wrap) {
 	case PIPE_TEX_WRAP_REPEAT:
-		ret = NV04_DX5_TEXTURED_TRIANGLE_FORMAT_ADDRESSU_REPEAT;
+		ret = NV04_TEXTURED_TRIANGLE_FORMAT_ADDRESSU_REPEAT;
 		break;
 	case PIPE_TEX_WRAP_MIRROR_REPEAT:
-		ret = NV04_DX5_TEXTURED_TRIANGLE_FORMAT_ADDRESSU_MIRRORED_REPEAT;
+		ret = NV04_TEXTURED_TRIANGLE_FORMAT_ADDRESSU_MIRRORED_REPEAT;
 		break;
 	case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
-		ret = NV04_DX5_TEXTURED_TRIANGLE_FORMAT_ADDRESSU_CLAMP_TO_EDGE;
+		ret = NV04_TEXTURED_TRIANGLE_FORMAT_ADDRESSU_CLAMP_TO_EDGE;
 		break;
 	case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
-		ret = NV04_DX5_TEXTURED_TRIANGLE_FORMAT_ADDRESSU_CLAMP_TO_BORDER;
+		ret = NV04_TEXTURED_TRIANGLE_FORMAT_ADDRESSU_CLAMP_TO_BORDER;
 		break;
 	case PIPE_TEX_WRAP_CLAMP:
-		ret = NV04_DX5_TEXTURED_TRIANGLE_FORMAT_ADDRESSU_CLAMP;
+		ret = NV04_TEXTURED_TRIANGLE_FORMAT_ADDRESSU_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 = NV04_DX5_TEXTURED_TRIANGLE_FORMAT_ADDRESSU_CLAMP;
+		ret = NV04_TEXTURED_TRIANGLE_FORMAT_ADDRESSU_CLAMP;
 	}
-	return ret >> NV04_DX5_TEXTURED_TRIANGLE_FORMAT_ADDRESSU_SHIFT;
+	return ret >> NV04_TEXTURED_TRIANGLE_FORMAT_ADDRESSU_SHIFT;
 }
 
 static void *
@@ -84,20 +84,20 @@ nv04_sampler_state_create(struct pipe_context *pipe,
 
 	ss = MALLOC(sizeof(struct nv04_sampler_state));
 
-	ss->format = ((wrap_mode(cso->wrap_s) << NV04_DX5_TEXTURED_TRIANGLE_FORMAT_ADDRESSU_SHIFT) |
-		    (wrap_mode(cso->wrap_t) << NV04_DX5_TEXTURED_TRIANGLE_FORMAT_ADDRESSV_SHIFT));
+	ss->format = ((wrap_mode(cso->wrap_s) << NV04_TEXTURED_TRIANGLE_FORMAT_ADDRESSU_SHIFT) |
+		    (wrap_mode(cso->wrap_t) << NV04_TEXTURED_TRIANGLE_FORMAT_ADDRESSV_SHIFT));
 
 	if (cso->max_anisotropy > 1.0) {
-		filter |= NV04_DX5_TEXTURED_TRIANGLE_FILTER_ANISOTROPIC_MINIFY_ENABLE | NV04_DX5_TEXTURED_TRIANGLE_FILTER_ANISOTROPIC_MAGNIFY_ENABLE;
+		filter |= NV04_TEXTURED_TRIANGLE_FILTER_ANISOTROPIC_MINIFY_ENABLE | NV04_TEXTURED_TRIANGLE_FILTER_ANISOTROPIC_MAGNIFY_ENABLE;
 	}
 
 	switch (cso->mag_img_filter) {
 	case PIPE_TEX_FILTER_LINEAR:
-		filter |= NV04_DX5_TEXTURED_TRIANGLE_FILTER_MAGNIFY_LINEAR;
+		filter |= NV04_TEXTURED_TRIANGLE_FILTER_MAGNIFY_LINEAR;
 		break;
 	case PIPE_TEX_FILTER_NEAREST:
 	default:
-		filter |= NV04_DX5_TEXTURED_TRIANGLE_FILTER_MAGNIFY_NEAREST;
+		filter |= NV04_TEXTURED_TRIANGLE_FILTER_MAGNIFY_NEAREST;
 		break;
 	}
 
@@ -105,14 +105,14 @@ nv04_sampler_state_create(struct pipe_context *pipe,
 	case PIPE_TEX_FILTER_LINEAR:
 		switch (cso->min_mip_filter) {
 		case PIPE_TEX_MIPFILTER_NEAREST:
-			filter |= NV04_DX5_TEXTURED_TRIANGLE_FILTER_MINIFY_LINEAR_MIPMAP_NEAREST;
+			filter |= NV04_TEXTURED_TRIANGLE_FILTER_MINIFY_LINEAR_MIPMAP_NEAREST;
 			break;
 		case PIPE_TEX_MIPFILTER_LINEAR:
-			filter |= NV04_DX5_TEXTURED_TRIANGLE_FILTER_MINIFY_LINEAR_MIPMAP_LINEAR;
+			filter |= NV04_TEXTURED_TRIANGLE_FILTER_MINIFY_LINEAR_MIPMAP_LINEAR;
 			break;
 		case PIPE_TEX_MIPFILTER_NONE:
 		default:
-			filter |= NV04_DX5_TEXTURED_TRIANGLE_FILTER_MINIFY_LINEAR;
+			filter |= NV04_TEXTURED_TRIANGLE_FILTER_MINIFY_LINEAR;
 			break;
 		}
 		break;
@@ -120,14 +120,14 @@ nv04_sampler_state_create(struct pipe_context *pipe,
 	default:
 		switch (cso->min_mip_filter) {
 		case PIPE_TEX_MIPFILTER_NEAREST:
-			filter |= NV04_DX5_TEXTURED_TRIANGLE_FILTER_MINIFY_NEAREST_MIPMAP_NEAREST;
+			filter |= NV04_TEXTURED_TRIANGLE_FILTER_MINIFY_NEAREST_MIPMAP_NEAREST;
 		break;
 		case PIPE_TEX_MIPFILTER_LINEAR:
-			filter |= NV04_DX5_TEXTURED_TRIANGLE_FILTER_MINIFY_NEAREST_MIPMAP_LINEAR;
+			filter |= NV04_TEXTURED_TRIANGLE_FILTER_MINIFY_NEAREST_MIPMAP_LINEAR;
 			break;
 		case PIPE_TEX_MIPFILTER_NONE:
 		default:
-			filter |= NV04_DX5_TEXTURED_TRIANGLE_FILTER_MINIFY_NEAREST;
+			filter |= NV04_TEXTURED_TRIANGLE_FILTER_MINIFY_NEAREST;
 			break;
 		}
 		break;
@@ -181,7 +181,7 @@ nv04_rasterizer_state_create(struct pipe_context *pipe,
 	 */
 	rs = MALLOC(sizeof(struct nv04_rasterizer_state));
 
-	rs->blend = cso->flatshade ? NV04_DX5_TEXTURED_TRIANGLE_BLEND_SHADE_MODE_FLAT : NV04_DX5_TEXTURED_TRIANGLE_BLEND_SHADE_MODE_GOURAUD;
+	rs->blend = cso->flatshade ? NV04_TEXTURED_TRIANGLE_BLEND_SHADE_MODE_FLAT : NV04_TEXTURED_TRIANGLE_BLEND_SHADE_MODE_GOURAUD;
 
 	return (void *)rs;
 }
@@ -229,16 +229,16 @@ nv04_depth_stencil_alpha_state_create(struct pipe_context *pipe,
 	hw = MALLOC(sizeof(struct nv04_depth_stencil_alpha_state));
 
 	hw->control = float_to_ubyte(cso->alpha.ref_value);
-	hw->control |= ( nv04_compare_func(cso->alpha.func) << NV04_DX5_TEXTURED_TRIANGLE_CONTROL_ALPHA_FUNC_SHIFT );
-	hw->control |= cso->alpha.enabled ? NV04_DX5_TEXTURED_TRIANGLE_CONTROL_ALPHA_TEST_ENABLE : 0;
-	hw->control |= NV04_DX5_TEXTURED_TRIANGLE_CONTROL_ORIGIN;
-	hw->control |= cso->depth.enabled ? (1 << NV04_DX5_TEXTURED_TRIANGLE_CONTROL_Z_ENABLE_SHIFT) : 0;
-	hw->control |= ( nv04_compare_func(cso->depth.func)<< NV04_DX5_TEXTURED_TRIANGLE_CONTROL_Z_FUNC_SHIFT );
-	hw->control |= 1 << NV04_DX5_TEXTURED_TRIANGLE_CONTROL_CULL_MODE_SHIFT; // no culling, handled by the draw module
-	hw->control |= NV04_DX5_TEXTURED_TRIANGLE_CONTROL_DITHER_ENABLE;
-	hw->control |= NV04_DX5_TEXTURED_TRIANGLE_CONTROL_Z_PERSPECTIVE_ENABLE;
-	hw->control |= cso->depth.writemask ? (1 << NV04_DX5_TEXTURED_TRIANGLE_CONTROL_Z_WRITE_ENABLE_SHIFT) : 0;
-	hw->control |= 1 << NV04_DX5_TEXTURED_TRIANGLE_CONTROL_Z_FORMAT_SHIFT; // integer zbuffer format
+	hw->control |= ( nv04_compare_func(cso->alpha.func) << NV04_TEXTURED_TRIANGLE_CONTROL_ALPHA_FUNC_SHIFT );
+	hw->control |= cso->alpha.enabled ? NV04_TEXTURED_TRIANGLE_CONTROL_ALPHA_ENABLE : 0;
+	hw->control |= NV04_TEXTURED_TRIANGLE_CONTROL_ORIGIN;
+	hw->control |= cso->depth.enabled ? NV04_TEXTURED_TRIANGLE_CONTROL_Z_ENABLE : 0;
+	hw->control |= ( nv04_compare_func(cso->depth.func)<< NV04_TEXTURED_TRIANGLE_CONTROL_Z_FUNC_SHIFT );
+	hw->control |= 1 << NV04_TEXTURED_TRIANGLE_CONTROL_CULL_MODE_SHIFT; // no culling, handled by the draw module
+	hw->control |= NV04_TEXTURED_TRIANGLE_CONTROL_DITHER_ENABLE;
+	hw->control |= NV04_TEXTURED_TRIANGLE_CONTROL_Z_PERSPECTIVE_ENABLE;
+	hw->control |= cso->depth.writemask ? NV04_TEXTURED_TRIANGLE_CONTROL_Z_WRITE : 0;
+	hw->control |= 1 << NV04_TEXTURED_TRIANGLE_CONTROL_Z_FORMAT_SHIFT; // integer zbuffer format
 
 	return (void *)hw;
 }
@@ -377,7 +377,7 @@ nv04_set_scissor_state(struct pipe_context *pipe,
 /*	struct nv04_context *nv04 = nv04_context(pipe);
 
 	// XXX
-	BEGIN_RING(fahrenheit, NV04_DX5_TEXTURED_TRIANGLE_SCISSOR_HORIZ, 2);
+	BEGIN_RING(fahrenheit, NV04_TEXTURED_TRIANGLE_SCISSOR_HORIZ, 2);
 	OUT_RING  (((s->maxx - s->minx) << 16) | s->minx);
 	OUT_RING  (((s->maxy - s->miny) << 16) | s->miny);*/
 }
@@ -425,9 +425,9 @@ nv04_init_state_functions(struct nv04_context *nv04)
 	nv04->pipe.delete_blend_state = nv04_blend_state_delete;
 
 	nv04->pipe.create_sampler_state = nv04_sampler_state_create;
-	nv04->pipe.bind_sampler_states = nv04_sampler_state_bind;
+	nv04->pipe.bind_fragment_sampler_states = nv04_sampler_state_bind;
 	nv04->pipe.delete_sampler_state = nv04_sampler_state_delete;
-	nv04->pipe.set_sampler_textures = nv04_set_sampler_texture;
+	nv04->pipe.set_fragment_sampler_textures = nv04_set_sampler_texture;
 
 	nv04->pipe.create_rasterizer_state = nv04_rasterizer_state_create;
 	nv04->pipe.bind_rasterizer_state = nv04_rasterizer_state_bind;
diff --git a/src/gallium/drivers/nv04/nv04_state.h b/src/gallium/drivers/nv04/nv04_state.h
index 399f750dbe..81d1d2ebaa 100644
--- a/src/gallium/drivers/nv04/nv04_state.h
+++ b/src/gallium/drivers/nv04/nv04_state.h
@@ -31,6 +31,7 @@ struct nv04_rasterizer_state {
 
 struct nv04_miptree {
 	struct pipe_texture base;
+	struct nouveau_bo *bo;
 
 	struct pipe_buffer *buffer;
 	uint total_size;
diff --git a/src/gallium/drivers/nv04/nv04_state_emit.c b/src/gallium/drivers/nv04/nv04_state_emit.c
index eb2c1c57c6..b8d6dc560f 100644
--- a/src/gallium/drivers/nv04/nv04_state_emit.c
+++ b/src/gallium/drivers/nv04/nv04_state_emit.c
@@ -57,13 +57,19 @@ static uint32_t nv04_blend_func(uint32_t f)
 static void nv04_emit_control(struct nv04_context* nv04)
 {
 	uint32_t control = nv04->dsa->control;
+	struct nv04_screen *screen = nv04->screen;
+	struct nouveau_channel *chan = screen->base.channel;
+	struct nouveau_grobj *fahrenheit = screen->fahrenheit;
 
-	BEGIN_RING(fahrenheit, NV04_DX5_TEXTURED_TRIANGLE_CONTROL, 1);
-	OUT_RING(control);
+	BEGIN_RING(chan, fahrenheit, NV04_TEXTURED_TRIANGLE_CONTROL, 1);
+	OUT_RING(chan, control);
 }
 
 static void nv04_emit_blend(struct nv04_context* nv04)
 {
+	struct nv04_screen *screen = nv04->screen;
+	struct nouveau_channel *chan = screen->base.channel;
+	struct nouveau_grobj *fahrenheit = screen->fahrenheit;
 	uint32_t blend;
 
 	blend=0x4; // texture MODULATE_ALPHA
@@ -75,19 +81,23 @@ static void nv04_emit_blend(struct nv04_context* nv04)
 	blend|=(nv04_blend_func(nv04->blend->b_src)<<24);
 	blend|=(nv04_blend_func(nv04->blend->b_dst)<<28);
 
-	BEGIN_RING(fahrenheit, NV04_DX5_TEXTURED_TRIANGLE_BLEND, 1);
-	OUT_RING(blend);
+	BEGIN_RING(chan, fahrenheit, NV04_TEXTURED_TRIANGLE_BLEND, 1);
+	OUT_RING(chan, blend);
 }
 
 static void nv04_emit_sampler(struct nv04_context *nv04, int unit)
 {
 	struct nv04_miptree *nv04mt = nv04->tex_miptree[unit];
 	struct pipe_texture *pt = &nv04mt->base;
-
-	BEGIN_RING(fahrenheit, NV04_DX5_TEXTURED_TRIANGLE_OFFSET, 3);
-	OUT_RELOCl(nv04mt->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD);
-	OUT_RELOCd(nv04mt->buffer, (nv04->fragtex.format | nv04->sampler[unit]->format), NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_OR | NOUVEAU_BO_RD, 1/*VRAM*/,2/*TT*/);
-	OUT_RING(nv04->sampler[unit]->filter);
+	struct nv04_screen *screen = nv04->screen;
+	struct nouveau_channel *chan = screen->base.channel;
+	struct nouveau_grobj *fahrenheit = screen->fahrenheit;
+	struct nouveau_bo *bo = nouveau_bo(nv04mt->buffer);
+
+	BEGIN_RING(chan, fahrenheit, NV04_TEXTURED_TRIANGLE_OFFSET, 3);
+	OUT_RELOCl(chan, bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD);
+	OUT_RELOCd(chan, bo, (nv04->fragtex.format | nv04->sampler[unit]->format), NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_OR | NOUVEAU_BO_RD, 1/*VRAM*/,2/*TT*/);
+	OUT_RING(chan, nv04->sampler[unit]->filter);
 }
 
 static void nv04_state_emit_framebuffer(struct nv04_context* nv04)
@@ -97,6 +107,10 @@ static void nv04_state_emit_framebuffer(struct nv04_context* nv04)
 	uint32_t rt_format, w, h;
 	int colour_format = 0, zeta_format = 0;
 	struct nv04_miptree *nv04mt = 0;
+	struct nv04_screen *screen = nv04->screen;
+	struct nouveau_channel *chan = screen->base.channel;
+	struct nouveau_grobj *context_surfaces_3d = screen->context_surfaces_3d;
+	struct nouveau_bo *bo;
 
 	w = fb->cbufs[0]->width;
 	h = fb->cbufs[0]->height;
@@ -128,24 +142,29 @@ static void nv04_state_emit_framebuffer(struct nv04_context* nv04)
 		assert(0);
 	}
 
-	BEGIN_RING(context_surfaces_3d, NV04_CONTEXT_SURFACES_3D_FORMAT, 1);
-	OUT_RING(rt_format);
+	BEGIN_RING(chan, context_surfaces_3d, NV04_CONTEXT_SURFACES_3D_FORMAT, 1);
+	OUT_RING(chan, rt_format);
 
 	nv04mt = (struct nv04_miptree *)rt->base.texture;
+	bo = nouveau_bo(nv04mt->buffer);
 	/* FIXME pitches have to be aligned ! */
-	BEGIN_RING(context_surfaces_3d, NV04_CONTEXT_SURFACES_3D_PITCH, 2);
-	OUT_RING(rt->pitch|(zeta->pitch<<16));
-	OUT_RELOCl(nv04mt->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
+	BEGIN_RING(chan, context_surfaces_3d, NV04_CONTEXT_SURFACES_3D_PITCH, 2);
+	OUT_RING(chan, rt->pitch|(zeta->pitch<<16));
+	OUT_RELOCl(chan, bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
 	if (fb->zsbuf) {
 		nv04mt = (struct nv04_miptree *)zeta->base.texture;
-		BEGIN_RING(context_surfaces_3d, NV04_CONTEXT_SURFACES_3D_OFFSET_ZETA, 1);
-		OUT_RELOCl(nv04mt->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
+		BEGIN_RING(chan, context_surfaces_3d, NV04_CONTEXT_SURFACES_3D_OFFSET_ZETA, 1);
+		OUT_RELOCl(chan, bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
 	}
 }
 
 void
 nv04_emit_hw_state(struct nv04_context *nv04)
 {
+	struct nv04_screen *screen = nv04->screen;
+	struct nouveau_channel *chan = screen->base.channel;
+	struct nouveau_grobj *fahrenheit = screen->fahrenheit;
+	struct nouveau_grobj *context_surfaces_3d = screen->context_surfaces_3d;
 	int i;
 
 	if (nv04->dirty & NV04_NEW_VERTPROG) {
@@ -163,8 +182,8 @@ nv04_emit_hw_state(struct nv04_context *nv04)
 	if (nv04->dirty & NV04_NEW_CONTROL) {
 		nv04->dirty &= ~NV04_NEW_CONTROL;
 
-		BEGIN_RING(fahrenheit, NV04_DX5_TEXTURED_TRIANGLE_CONTROL, 1);
-		OUT_RING(nv04->dsa->control);
+		BEGIN_RING(chan, fahrenheit, NV04_TEXTURED_TRIANGLE_CONTROL, 1);
+		OUT_RING(chan, nv04->dsa->control);
 	}
 
 	if (nv04->dirty & NV04_NEW_BLEND) {
@@ -205,12 +224,12 @@ nv04_emit_hw_state(struct nv04_context *nv04)
 	unsigned rt_pitch = ((struct nv04_surface *)nv04->rt)->pitch;
 	unsigned zeta_pitch = ((struct nv04_surface *)nv04->zeta)->pitch;
 
-	BEGIN_RING(context_surfaces_3d, NV04_CONTEXT_SURFACES_3D_PITCH, 2);
-	OUT_RING(rt_pitch|(zeta_pitch<<16));
-	OUT_RELOCl(nv04->rt, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
+	BEGIN_RING(chan, context_surfaces_3d, NV04_CONTEXT_SURFACES_3D_PITCH, 2);
+	OUT_RING(chan, rt_pitch|(zeta_pitch<<16));
+	OUT_RELOCl(chan, nouveau_bo(nv04->rt), 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
 	if (nv04->zeta) {
-		BEGIN_RING(context_surfaces_3d, NV04_CONTEXT_SURFACES_3D_OFFSET_ZETA, 1);
-		OUT_RELOCl(nv04->zeta, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
+		BEGIN_RING(chan, context_surfaces_3d, NV04_CONTEXT_SURFACES_3D_OFFSET_ZETA, 1);
+		OUT_RELOCl(chan, nouveau_bo(nv04->zeta), 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
 	}
 
 	/* Texture images */
@@ -218,9 +237,10 @@ nv04_emit_hw_state(struct nv04_context *nv04)
 		if (!(nv04->fp_samplers & (1 << i)))
 			continue;
 		struct nv04_miptree *nv04mt = nv04->tex_miptree[i];
-		BEGIN_RING(fahrenheit, NV04_DX5_TEXTURED_TRIANGLE_OFFSET, 2);
-		OUT_RELOCl(nv04mt->buffer, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD);
-		OUT_RELOCd(nv04mt->buffer, (nv04->fragtex.format | nv04->sampler[i]->format), NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_OR | NOUVEAU_BO_RD, 1/*VRAM*/,2/*TT*/);
+		struct nouveau_bo *bo = nouveau_bo(nv04mt->buffer);
+		BEGIN_RING(chan, fahrenheit, NV04_TEXTURED_TRIANGLE_OFFSET, 2);
+		OUT_RELOCl(chan, bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD);
+		OUT_RELOCd(chan, bo, (nv04->fragtex.format | nv04->sampler[i]->format), NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_OR | NOUVEAU_BO_RD, 1/*VRAM*/,2/*TT*/);
 	}
 }
 
diff --git a/src/gallium/drivers/nv04/nv04_surface_2d.c b/src/gallium/drivers/nv04/nv04_surface_2d.c
index b2ab50ee21..b24a9cee5a 100644
--- a/src/gallium/drivers/nv04/nv04_surface_2d.c
+++ b/src/gallium/drivers/nv04/nv04_surface_2d.c
@@ -1,5 +1,6 @@
 #include "pipe/p_context.h"
 #include "pipe/p_format.h"
+#include "util/u_format.h"
 #include "util/u_math.h"
 #include "util/u_memory.h"
 
@@ -13,10 +14,13 @@ nv04_surface_format(enum pipe_format format)
 {
 	switch (format) {
 	case PIPE_FORMAT_A8_UNORM:
+	case PIPE_FORMAT_L8_UNORM:
+	case PIPE_FORMAT_I8_UNORM:
 		return NV04_CONTEXT_SURFACES_2D_FORMAT_Y8;
 	case PIPE_FORMAT_R16_SNORM:
 	case PIPE_FORMAT_R5G6B5_UNORM:
 	case PIPE_FORMAT_Z16_UNORM:
+	case PIPE_FORMAT_A8L8_UNORM:
 		return NV04_CONTEXT_SURFACES_2D_FORMAT_R5G6B5;
 	case PIPE_FORMAT_X8R8G8B8_UNORM:
 	case PIPE_FORMAT_A8R8G8B8_UNORM:
@@ -36,8 +40,10 @@ nv04_rect_format(enum pipe_format format)
 	case PIPE_FORMAT_A8_UNORM:
 		return NV04_GDI_RECTANGLE_TEXT_COLOR_FORMAT_A8R8G8B8;
 	case PIPE_FORMAT_R5G6B5_UNORM:
+	case PIPE_FORMAT_A8L8_UNORM:
 	case PIPE_FORMAT_Z16_UNORM:
 		return NV04_GDI_RECTANGLE_TEXT_COLOR_FORMAT_A16R5G6B5;
+	case PIPE_FORMAT_X8R8G8B8_UNORM:
 	case PIPE_FORMAT_A8R8G8B8_UNORM:
 	case PIPE_FORMAT_Z24S8_UNORM:
 	case PIPE_FORMAT_Z24X8_UNORM:
@@ -51,6 +57,10 @@ static INLINE int
 nv04_scaled_image_format(enum pipe_format format)
 {
 	switch (format) {
+	case PIPE_FORMAT_A8_UNORM:
+	case PIPE_FORMAT_L8_UNORM:
+	case PIPE_FORMAT_I8_UNORM:
+		return NV04_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_Y8;
 	case PIPE_FORMAT_A1R5G5B5_UNORM:
 		return NV04_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_A1R5G5B5;
 	case PIPE_FORMAT_A8R8G8B8_UNORM:
@@ -59,6 +69,7 @@ nv04_scaled_image_format(enum pipe_format format)
 		return NV04_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_X8R8G8B8;
 	case PIPE_FORMAT_R5G6B5_UNORM:
 	case PIPE_FORMAT_R16_SNORM:
+	case PIPE_FORMAT_A8L8_UNORM:
 		return NV04_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_R5G6B5;
 	default:
 		return -1;
@@ -66,7 +77,7 @@ nv04_scaled_image_format(enum pipe_format format)
 }
 
 static INLINE unsigned
-nv04_swizzle_bits(unsigned x, unsigned y)
+nv04_swizzle_bits_square(unsigned x, unsigned y)
 {
 	unsigned u = (x & 0x001) << 0 |
 	             (x & 0x002) << 1 |
@@ -96,6 +107,15 @@ nv04_swizzle_bits(unsigned x, unsigned y)
 	return v | u;
 }
 
+/* rectangular swizzled textures are linear concatenations of swizzled square tiles */
+static INLINE unsigned
+nv04_swizzle_bits(unsigned x, unsigned y, unsigned w, unsigned h)
+{
+	unsigned s = MIN2(w, h);
+	unsigned m = s - 1;
+	return (((x | y) & ~m) * s) | nv04_swizzle_bits_square(x & m, y & m);
+}
+
 static int
 nv04_surface_copy_swizzle(struct nv04_surface_2d *ctx,
 			  struct pipe_surface *dst, int dx, int dy,
@@ -123,6 +143,9 @@ nv04_surface_copy_swizzle(struct nv04_surface_2d *ctx,
 	assert(sub_w == w || util_is_pot(sub_w));
 	assert(sub_h == h || util_is_pot(sub_h));
 
+	MARK_RING (chan, 8 + ((w+sub_w)/sub_w)*((h+sub_h)/sub_h)*17, 2 +
+			 ((w+sub_w)/sub_w)*((h+sub_h)/sub_h)*2);
+
 	BEGIN_RING(chan, swzsurf, NV04_SWIZZLED_SURFACE_DMA_IMAGE, 1);
 	OUT_RELOCo(chan, dst_bo,
 	                 NOUVEAU_BO_GART | NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
@@ -131,7 +154,7 @@ nv04_surface_copy_swizzle(struct nv04_surface_2d *ctx,
 	OUT_RING  (chan, nv04_surface_format(dst->format) |
 	                 log2i(dst->width) << NV04_SWIZZLED_SURFACE_FORMAT_BASE_SIZE_U_SHIFT |
 	                 log2i(dst->height) << NV04_SWIZZLED_SURFACE_FORMAT_BASE_SIZE_V_SHIFT);
- 
+
 	BEGIN_RING(chan, sifm, NV04_SCALED_IMAGE_FROM_MEMORY_DMA_IMAGE, 1);
 	OUT_RELOCo(chan, src_bo,
 	                 NOUVEAU_BO_GART | NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
@@ -144,20 +167,19 @@ nv04_surface_copy_swizzle(struct nv04_surface_2d *ctx,
 	  for (x = 0; x < w; x += sub_w) {
 	    sub_w = MIN2(sub_w, w - x);
 
-	    /* Must be 64-byte aligned */
-	    assert(!((dst->offset + nv04_swizzle_bits(dx+x, dy+y) * dst->texture->block.size) & 63));
+	    assert(!(dst->offset & 63));
 
 	    BEGIN_RING(chan, swzsurf, NV04_SWIZZLED_SURFACE_OFFSET, 1);
-	    OUT_RELOCl(chan, dst_bo, dst->offset + nv04_swizzle_bits(dx+x, dy+y) * dst->texture->block.size,
+	    OUT_RELOCl(chan, dst_bo, dst->offset,
                              NOUVEAU_BO_GART | NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
 
 	    BEGIN_RING(chan, sifm, NV04_SCALED_IMAGE_FROM_MEMORY_COLOR_CONVERSION, 9);
 	    OUT_RING  (chan, NV04_SCALED_IMAGE_FROM_MEMORY_COLOR_CONVERSION_TRUNCATE);
 	    OUT_RING  (chan, nv04_scaled_image_format(src->format));
 	    OUT_RING  (chan, NV04_SCALED_IMAGE_FROM_MEMORY_OPERATION_SRCCOPY);
-	    OUT_RING  (chan, 0);
+	    OUT_RING  (chan, (x + dx) | ((y + dy) << NV04_SCALED_IMAGE_FROM_MEMORY_CLIP_POINT_Y_SHIFT));
 	    OUT_RING  (chan, sub_h << NV04_SCALED_IMAGE_FROM_MEMORY_CLIP_SIZE_H_SHIFT | sub_w);
-	    OUT_RING  (chan, 0);
+	    OUT_RING  (chan, (x + dx) | ((y + dy) << NV04_SCALED_IMAGE_FROM_MEMORY_OUT_POINT_Y_SHIFT));
 	    OUT_RING  (chan, sub_h << NV04_SCALED_IMAGE_FROM_MEMORY_OUT_SIZE_H_SHIFT | sub_w);
 	    OUT_RING  (chan, 1 << 20);
 	    OUT_RING  (chan, 1 << 20);
@@ -167,7 +189,7 @@ nv04_surface_copy_swizzle(struct nv04_surface_2d *ctx,
 	    OUT_RING  (chan, src_pitch |
 			     NV04_SCALED_IMAGE_FROM_MEMORY_FORMAT_ORIGIN_CENTER |
 			     NV04_SCALED_IMAGE_FROM_MEMORY_FORMAT_FILTER_POINT_SAMPLE);
-	    OUT_RELOCl(chan, src_bo, src->offset + (sy+y) * src_pitch + (sx+x) * src->texture->block.size,
+	    OUT_RELOCl(chan, src_bo, src->offset + (sy+y) * src_pitch + (sx+x) * util_format_get_blocksize(src->texture->format),
                              NOUVEAU_BO_GART | NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
 	    OUT_RING  (chan, 0);
 	  }
@@ -188,11 +210,11 @@ nv04_surface_copy_m2mf(struct nv04_surface_2d *ctx,
 	unsigned src_pitch = ((struct nv04_surface *)src)->pitch;
 	unsigned dst_pitch = ((struct nv04_surface *)dst)->pitch;
 	unsigned dst_offset = dst->offset + dy * dst_pitch +
-	                      dx * dst->texture->block.size;
+	                      dx * util_format_get_blocksize(dst->texture->format);
 	unsigned src_offset = src->offset + sy * src_pitch +
-	                      sx * src->texture->block.size;
+	                      sx * util_format_get_blocksize(src->texture->format);
 
-	WAIT_RING (chan, 3 + ((h / 2047) + 1) * 9);
+	MARK_RING (chan, 3 + ((h / 2047) + 1) * 9, 2 + ((h / 2047) + 1) * 2);
 	BEGIN_RING(chan, m2mf, NV04_MEMORY_TO_MEMORY_FORMAT_DMA_BUFFER_IN, 2);
 	OUT_RELOCo(chan, src_bo,
 		   NOUVEAU_BO_GART | NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
@@ -209,7 +231,7 @@ nv04_surface_copy_m2mf(struct nv04_surface_2d *ctx,
 			   NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_WR);
 		OUT_RING  (chan, src_pitch);
 		OUT_RING  (chan, dst_pitch);
-		OUT_RING  (chan, w * src->texture->block.size);
+		OUT_RING  (chan, w * util_format_get_blocksize(src->texture->format));
 		OUT_RING  (chan, count);
 		OUT_RING  (chan, 0x0101);
 		OUT_RING  (chan, 0);
@@ -240,7 +262,7 @@ nv04_surface_copy_blit(struct nv04_surface_2d *ctx, struct pipe_surface *dst,
 	if (format < 0)
 		return 1;
 
-	WAIT_RING (chan, 12);
+	MARK_RING (chan, 12, 4);
 	BEGIN_RING(chan, surf2d, NV04_CONTEXT_SURFACES_2D_DMA_IMAGE_SOURCE, 2);
 	OUT_RELOCo(chan, src_bo, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
 	OUT_RELOCo(chan, dst_bo, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
@@ -305,7 +327,7 @@ nv04_surface_fill(struct nv04_surface_2d *ctx, struct pipe_surface *dst,
 	gdirect_format = nv04_rect_format(dst->format);
 	assert(gdirect_format >= 0);
 
-	WAIT_RING (chan, 16);
+	MARK_RING (chan, 16, 4);
 	BEGIN_RING(chan, surf2d, NV04_CONTEXT_SURFACES_2D_DMA_IMAGE_SOURCE, 2);
 	OUT_RELOCo(chan, dst_bo, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
 	OUT_RELOCo(chan, dst_bo, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
@@ -477,3 +499,49 @@ nv04_surface_2d_init(struct nouveau_screen *screen)
 	ctx->fill = nv04_surface_fill;
 	return ctx;
 }
+
+struct nv04_surface*
+nv04_surface_wrap_for_render(struct pipe_screen *pscreen, struct nv04_surface_2d* eng2d, struct nv04_surface* ns)
+{
+	int temp_flags;
+
+	// printf("creating temp, flags is %i!\n", flags);
+
+	if(ns->base.usage & PIPE_BUFFER_USAGE_DISCARD)
+	{
+		temp_flags = ns->base.usage | PIPE_BUFFER_USAGE_GPU_READ;
+		ns->base.usage = PIPE_BUFFER_USAGE_GPU_WRITE | NOUVEAU_BUFFER_USAGE_NO_RENDER | PIPE_BUFFER_USAGE_DISCARD;
+	}
+	else
+	{
+		temp_flags = ns->base.usage | PIPE_BUFFER_USAGE_GPU_READ | PIPE_BUFFER_USAGE_GPU_WRITE;
+		ns->base.usage = PIPE_BUFFER_USAGE_GPU_WRITE | NOUVEAU_BUFFER_USAGE_NO_RENDER | PIPE_BUFFER_USAGE_GPU_READ;
+	}
+
+	struct nv40_screen* screen = (struct nv40_screen*)pscreen;
+	ns->base.usage = PIPE_BUFFER_USAGE_GPU_READ | PIPE_BUFFER_USAGE_GPU_WRITE;
+
+	struct pipe_texture templ;
+	memset(&templ, 0, sizeof(templ));
+	templ.format = ns->base.texture->format;
+	templ.target = PIPE_TEXTURE_2D;
+	templ.width0 = ns->base.width;
+	templ.height0 = ns->base.height;
+	templ.depth0 = 1;
+	templ.last_level = 0;
+
+	// TODO: this is probably wrong and we should specifically handle multisampling somehow once it is implemented
+	templ.nr_samples = ns->base.texture->nr_samples;
+
+	templ.tex_usage = ns->base.texture->tex_usage | PIPE_TEXTURE_USAGE_RENDER_TARGET;
+
+	struct pipe_texture* temp_tex = pscreen->texture_create(pscreen, &templ);
+	struct nv04_surface* temp_ns = (struct nv04_surface*)pscreen->get_tex_surface(pscreen, temp_tex, 0, 0, 0, temp_flags);
+	temp_ns->backing = ns;
+
+	if(ns->base.usage & PIPE_BUFFER_USAGE_GPU_READ)
+		eng2d->copy(eng2d, &temp_ns->backing->base, 0, 0, &ns->base, 0, 0, ns->base.width, ns->base.height);
+
+	return temp_ns;
+}
+
diff --git a/src/gallium/drivers/nv04/nv04_surface_2d.h b/src/gallium/drivers/nv04/nv04_surface_2d.h
index 02b3f56ba8..ce696a11a3 100644
--- a/src/gallium/drivers/nv04/nv04_surface_2d.h
+++ b/src/gallium/drivers/nv04/nv04_surface_2d.h
@@ -4,6 +4,7 @@
 struct nv04_surface {
 	struct pipe_surface base;
 	unsigned pitch;
+	struct nv04_surface* backing;
 };
 
 struct nv04_surface_2d {
@@ -30,4 +31,7 @@ nv04_surface_2d_init(struct nouveau_screen *screen);
 void
 nv04_surface_2d_takedown(struct nv04_surface_2d **);
 
+struct nv04_surface*
+nv04_surface_wrap_for_render(struct pipe_screen *pscreen, struct nv04_surface_2d* eng2d, struct nv04_surface* ns);
+
 #endif
diff --git a/src/gallium/drivers/nv04/nv04_transfer.c b/src/gallium/drivers/nv04/nv04_transfer.c
index 6618660743..2dd2e146a8 100644
--- a/src/gallium/drivers/nv04/nv04_transfer.c
+++ b/src/gallium/drivers/nv04/nv04_transfer.c
@@ -1,7 +1,9 @@
 #include <pipe/p_state.h>
 #include <pipe/p_defines.h>
 #include <pipe/p_inlines.h>
+#include <util/u_format.h>
 #include <util/u_memory.h>
+#include <util/u_math.h>
 #include <nouveau/nouveau_winsys.h>
 #include "nv04_context.h"
 #include "nv04_screen.h"
@@ -10,22 +12,19 @@
 struct nv04_transfer {
 	struct pipe_transfer base;
 	struct pipe_surface *surface;
-	bool direct;
+	boolean direct;
 };
 
 static void
-nv04_compatible_transfer_tex(struct pipe_texture *pt, unsigned level,
+nv04_compatible_transfer_tex(struct pipe_texture *pt, unsigned width, unsigned height,
                              struct pipe_texture *template)
 {
 	memset(template, 0, sizeof(struct pipe_texture));
 	template->target = pt->target;
 	template->format = pt->format;
-	template->width[0] = pt->width[level];
-	template->height[0] = pt->height[level];
-	template->depth[0] = 1;
-	template->block = pt->block;
-	template->nblocksx[0] = pt->nblocksx[level];
-	template->nblocksy[0] = pt->nblocksx[level];
+	template->width0 = width;
+	template->height0 = height;
+	template->depth0 = 1;
 	template->last_level = 0;
 	template->nr_samples = pt->nr_samples;
 
@@ -48,14 +47,10 @@ nv04_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 		return NULL;
 
 	pipe_texture_reference(&tx->base.texture, pt);
-	tx->base.format = pt->format;
 	tx->base.x = x;
 	tx->base.y = y;
 	tx->base.width = w;
 	tx->base.height = h;
-	tx->base.block = pt->block;
-	tx->base.nblocksx = pt->nblocksx[level];
-	tx->base.nblocksy = pt->nblocksy[level];
 	tx->base.stride = mt->level[level].pitch;
 	tx->base.usage = usage;
 	tx->base.face = face;
@@ -76,7 +71,7 @@ nv04_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 
 	tx->direct = false;
 
-	nv04_compatible_transfer_tex(pt, level, &tx_tex_template);
+	nv04_compatible_transfer_tex(pt, w, h, &tx_tex_template);
 
 	tx_tex = pscreen->texture_create(pscreen, &tx_tex_template);
 	if (!tx_tex)
@@ -85,6 +80,8 @@ nv04_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 		return NULL;
 	}
 
+	tx->base.stride = ((struct nv04_miptree*)tx_tex)->level[0].pitch;
+
 	tx->surface = pscreen->get_tex_surface(pscreen, tx_tex,
 	                                       face, level, zslice,
 	                                       pipe_transfer_buffer_flags(&tx->base));
@@ -110,8 +107,8 @@ nv04_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 		/* TODO: Check if SIFM can un-swizzle */
 		nvscreen->eng2d->copy(nvscreen->eng2d,
 		                      tx->surface, 0, 0,
-		                      src, 0, 0,
-		                      src->width, src->height);
+		                      src, x, y,
+		                      w, h);
 
 		pipe_surface_reference(&src, NULL);
 	}
@@ -135,9 +132,9 @@ nv04_transfer_del(struct pipe_transfer *ptx)
 
 		/* TODO: Check if SIFM can deal with x,y,w,h when swizzling */
 		nvscreen->eng2d->copy(nvscreen->eng2d,
-		                      dst, 0, 0,
+		                      dst, tx->base.x, tx->base.y,
 		                      tx->surface, 0, 0,
-		                      dst->width, dst->height);
+		                      tx->base.width, tx->base.height);
 
 		pipe_surface_reference(&dst, NULL);
 	}
@@ -156,8 +153,10 @@ nv04_transfer_map(struct pipe_screen *pscreen, struct pipe_transfer *ptx)
 	void *map = pipe_buffer_map(pscreen, mt->buffer,
 	                            pipe_transfer_buffer_flags(ptx));
 
-	return map + ns->base.offset +
-	       ptx->y * ns->pitch + ptx->x * ptx->block.size;
+	if(!tx->direct)
+		return map + ns->base.offset;
+	else
+		return map + ns->base.offset + ptx->y * ns->pitch + ptx->x * util_format_get_blocksize(ptx->texture->format);
 }
 
 static void
diff --git a/src/gallium/drivers/nv04/nv04_vbo.c b/src/gallium/drivers/nv04/nv04_vbo.c
index e3167814f2..3484771814 100644
--- a/src/gallium/drivers/nv04/nv04_vbo.c
+++ b/src/gallium/drivers/nv04/nv04_vbo.c
@@ -9,7 +9,7 @@
 #include "nouveau/nouveau_channel.h"
 #include "nouveau/nouveau_pushbuf.h"
 
-boolean nv04_draw_elements( struct pipe_context *pipe,
+void nv04_draw_elements( struct pipe_context *pipe,
                     struct pipe_buffer *indexBuffer,
                     unsigned indexSize,
                     unsigned prim, unsigned start, unsigned count)
@@ -45,7 +45,7 @@ boolean nv04_draw_elements( struct pipe_context *pipe,
 		draw_set_mapped_element_buffer(draw, 0, NULL);
 	}
 
-	draw_set_mapped_constant_buffer(draw,
+	draw_set_mapped_constant_buffer(draw, PIPE_SHADER_VERTEX,
 					nv04->constbuf[PIPE_SHADER_VERTEX],
 					nv04->constbuf_nr[PIPE_SHADER_VERTEX]);
 
@@ -65,15 +65,13 @@ boolean nv04_draw_elements( struct pipe_context *pipe,
 		pipe_buffer_unmap(pscreen, indexBuffer);
 		draw_set_mapped_element_buffer(draw, 0, NULL);
 	}
-
-	return TRUE;
 }
 
-boolean nv04_draw_arrays( struct pipe_context *pipe,
-				 unsigned prim, unsigned start, unsigned count)
+void nv04_draw_arrays( struct pipe_context *pipe,
+                       unsigned prim, unsigned start, unsigned count)
 {
 	printf("coucou in draw arrays\n");
-	return nv04_draw_elements(pipe, NULL, 0, prim, start, count);
+	nv04_draw_elements(pipe, NULL, 0, prim, start, count);
 }