nv04-nv40: move 2d blit/fill code into pipe driver

19 files changed, 601 insertions, 42 deletions

diff --git a/src/gallium/drivers/nouveau/nouveau_winsys.h b/src/gallium/drivers/nouveau/nouveau_winsys.h
index 99f8e08201..b86c4b9338 100644
--- a/src/gallium/drivers/nouveau/nouveau_winsys.h
+++ b/src/gallium/drivers/nouveau/nouveau_winsys.h
@@ -50,7 +50,7 @@ struct nouveau_winsys {
 	uint32_t  (*notifier_status)(struct nouveau_notifier *, int id);
 	uint32_t  (*notifier_retval)(struct nouveau_notifier *, int id);
 	int       (*notifier_wait)(struct nouveau_notifier *, int id,
-				   int status, float timeout);
+				   int status, double timeout);
 
 	int (*surface_copy)(struct nouveau_winsys *, struct pipe_surface *,
 			    unsigned, unsigned, struct pipe_surface *,
diff --git a/src/gallium/drivers/nv04/Makefile b/src/gallium/drivers/nv04/Makefile
index 5ea51a2f42..4ed62dae95 100644
--- a/src/gallium/drivers/nv04/Makefile
+++ b/src/gallium/drivers/nv04/Makefile
@@ -4,6 +4,7 @@ include $(TOP)/configs/current
 LIBNAME = nv04
 
 DRIVER_SOURCES = \
+	nv04_surface_2d.c \
 	nv04_clear.c \
 	nv04_context.c \
 	nv04_fragprog.c \
diff --git a/src/gallium/drivers/nv04/nv04_screen.c b/src/gallium/drivers/nv04/nv04_screen.c
index e5e3d4772a..9ef38bc244 100644
--- a/src/gallium/drivers/nv04/nv04_screen.c
+++ b/src/gallium/drivers/nv04/nv04_screen.c
@@ -149,10 +149,19 @@ nv04_screen_destroy(struct pipe_screen *pscreen)
 
 	nvws->notifier_free(&screen->sync);
 	nvws->grobj_free(&screen->fahrenheit);
+	nv04_surface_2d_takedown(&screen->eng2d);
 
 	FREE(pscreen);
 }
 
+static struct pipe_buffer *
+nv04_surface_buffer(struct pipe_surface *surf)
+{
+	struct nv04_miptree *mt = (struct nv04_miptree *)surf->texture;
+
+	return mt->buffer;
+}
+
 struct pipe_screen *
 nv04_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws)
 {
@@ -181,6 +190,10 @@ nv04_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws)
 		return NULL;
 	}
 
+	/* 2D engine setup */
+	screen->eng2d = nv04_surface_2d_init(nvws);
+	screen->eng2d->buf = nv04_surface_buffer;
+
 	/* 3D object */
 	ret = nvws->grobj_alloc(nvws, fahrenheit_class, &screen->fahrenheit);
 	if (ret) {
diff --git a/src/gallium/drivers/nv04/nv04_screen.h b/src/gallium/drivers/nv04/nv04_screen.h
index 99a49cdf7a..540aec907b 100644
--- a/src/gallium/drivers/nv04/nv04_screen.h
+++ b/src/gallium/drivers/nv04/nv04_screen.h
@@ -2,6 +2,7 @@
 #define __NV04_SCREEN_H__
 
 #include "pipe/p_screen.h"
+#include "nv04_surface_2d.h"
 
 struct nv04_screen {
 	struct pipe_screen pipe;
@@ -10,6 +11,7 @@ struct nv04_screen {
 	unsigned chipset;
 
 	/* HW graphics objects */
+	struct nv04_surface_2d *eng2d;
 	struct nouveau_grobj *fahrenheit;
 	struct nouveau_grobj *context_surfaces_3d;
 	struct nouveau_notifier *sync;
diff --git a/src/gallium/drivers/nv04/nv04_surface.c b/src/gallium/drivers/nv04/nv04_surface.c
index 0d0983f9d4..1d11f53f2a 100644
--- a/src/gallium/drivers/nv04/nv04_surface.c
+++ b/src/gallium/drivers/nv04/nv04_surface.c
@@ -39,10 +39,17 @@ nv04_surface_copy(struct pipe_context *pipe, boolean do_flip,
 		  unsigned width, unsigned height)
 {
 	struct nv04_context *nv04 = nv04_context(pipe);
-	struct nouveau_winsys *nvws = nv04->nvws;
+	struct nv04_surface_2d *eng2d = nv04->screen->eng2d;
 
-	nvws->surface_copy(nvws, dest, destx, desty, src, srcx, srcy,
-			   width, height);
+	if (do_flip) {
+		desty += height;
+		while (height--) {
+			eng2d->copy(eng2d, dest, destx, desty--, src,
+				    srcx, srcy++, width, 1);
+		}
+	}
+
+	eng2d->copy(eng2d, dest, destx, desty, src, srcx, srcy, width, height);
 }
 
 static void
@@ -51,9 +58,9 @@ nv04_surface_fill(struct pipe_context *pipe, struct pipe_surface *dest,
 		  unsigned height, unsigned value)
 {
 	struct nv04_context *nv04 = nv04_context(pipe);
-	struct nouveau_winsys *nvws = nv04->nvws;
+	struct nv04_surface_2d *eng2d = nv04->screen->eng2d;
 
-	nvws->surface_fill(nvws, dest, destx, desty, width, height, value);
+	eng2d->fill(eng2d, dest, destx, desty, width, height, value);
 }
 
 void
diff --git a/src/gallium/drivers/nv04/nv04_surface_2d.c b/src/gallium/drivers/nv04/nv04_surface_2d.c
new file mode 100644
index 0000000000..7529583151
--- /dev/null
+++ b/src/gallium/drivers/nv04/nv04_surface_2d.c
@@ -0,0 +1,449 @@
+#include "pipe/p_context.h"
+#include "pipe/p_format.h"
+#include "util/u_memory.h"
+
+#include "nouveau/nouveau_winsys.h"
+#include "nouveau/nouveau_util.h"
+#include "nv04_surface_2d.h"
+
+static INLINE int
+nv04_surface_format(enum pipe_format format)
+{
+	switch (format) {
+	case PIPE_FORMAT_A8_UNORM:
+		return NV04_CONTEXT_SURFACES_2D_FORMAT_Y8;
+	case PIPE_FORMAT_R16_SNORM:
+	case PIPE_FORMAT_R5G6B5_UNORM:
+		return NV04_CONTEXT_SURFACES_2D_FORMAT_R5G6B5;
+	case PIPE_FORMAT_X8R8G8B8_UNORM:
+	case PIPE_FORMAT_A8R8G8B8_UNORM:
+		return NV04_CONTEXT_SURFACES_2D_FORMAT_A8R8G8B8;
+	case PIPE_FORMAT_Z24S8_UNORM:
+		return NV04_CONTEXT_SURFACES_2D_FORMAT_Y32;
+	default:
+		return -1;
+	}
+}
+
+static INLINE int
+nv04_rect_format(enum pipe_format format)
+{
+	switch (format) {
+	case PIPE_FORMAT_A8_UNORM:
+		return NV04_GDI_RECTANGLE_TEXT_COLOR_FORMAT_A8R8G8B8;
+	case PIPE_FORMAT_R5G6B5_UNORM:
+		return NV04_GDI_RECTANGLE_TEXT_COLOR_FORMAT_A16R5G6B5;
+	case PIPE_FORMAT_A8R8G8B8_UNORM:
+	case PIPE_FORMAT_Z24S8_UNORM:
+		return NV04_GDI_RECTANGLE_TEXT_COLOR_FORMAT_A8R8G8B8;
+	default:
+		return -1;
+	}
+}
+
+static INLINE int
+nv04_scaled_image_format(enum pipe_format format)
+{
+	switch (format) {
+	case PIPE_FORMAT_A1R5G5B5_UNORM:
+		return NV04_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_A1R5G5B5;
+	case PIPE_FORMAT_A8R8G8B8_UNORM:
+		return NV04_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_A8R8G8B8;
+	case PIPE_FORMAT_X8R8G8B8_UNORM:
+		return NV04_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_X8R8G8B8;
+	case PIPE_FORMAT_R5G6B5_UNORM:
+	case PIPE_FORMAT_R16_SNORM:
+		return NV04_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_R5G6B5;
+	default:
+		return -1;
+	}
+}
+
+static INLINE unsigned
+nv04_swizzle_bits(unsigned x, unsigned y)
+{
+	unsigned u = (x & 0x001) << 0 |
+	             (x & 0x002) << 1 |
+	             (x & 0x004) << 2 |
+	             (x & 0x008) << 3 |
+	             (x & 0x010) << 4 |
+	             (x & 0x020) << 5 |
+	             (x & 0x040) << 6 |
+	             (x & 0x080) << 7 |
+	             (x & 0x100) << 8 |
+	             (x & 0x200) << 9 |
+	             (x & 0x400) << 10 |
+	             (x & 0x800) << 11;
+
+	unsigned v = (y & 0x001) << 1 |
+	             (y & 0x002) << 2 |
+	             (y & 0x004) << 3 |
+	             (y & 0x008) << 4 |
+	             (y & 0x010) << 5 |
+	             (y & 0x020) << 6 |
+	             (y & 0x040) << 7 |
+	             (y & 0x080) << 8 |
+	             (y & 0x100) << 9 |
+	             (y & 0x200) << 10 |
+	             (y & 0x400) << 11 |
+	             (y & 0x800) << 12;
+	return v | u;
+}
+
+static int
+nv04_surface_copy_swizzle(struct nv04_surface_2d *ctx,
+			  struct pipe_surface *dst, int dx, int dy,
+			  struct pipe_surface *src, int sx, int sy,
+			  int w, int h)
+{
+	struct nouveau_channel *chan = ctx->nvws->channel;
+	struct nouveau_grobj *swzsurf = ctx->swzsurf;
+	struct nouveau_grobj *sifm = ctx->sifm;
+	struct nouveau_bo *src_bo = ctx->nvws->get_bo(ctx->buf(src));
+	struct nouveau_bo *dst_bo = ctx->nvws->get_bo(ctx->buf(dst));
+	const unsigned max_w = 1024;
+	const unsigned max_h = 1024;
+	const unsigned sub_w = w > max_w ? max_w : w;
+	const unsigned sub_h = h > max_h ? max_h : h;
+	unsigned cx = 0;
+	unsigned cy = 0;
+
+	/* POT or GTFO */
+	assert(!(w & (w - 1)) && !(h & (h - 1)));
+
+	BEGIN_RING(chan, swzsurf, NV04_SWIZZLED_SURFACE_DMA_IMAGE, 1);
+	OUT_RELOCo(chan, dst_bo,
+	                 NOUVEAU_BO_GART | NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
+	BEGIN_RING(chan, swzsurf, NV04_SWIZZLED_SURFACE_FORMAT, 1);
+	OUT_RING  (chan, nv04_surface_format(dst->format) |
+	                 log2i(w) << NV04_SWIZZLED_SURFACE_FORMAT_BASE_SIZE_U_SHIFT |
+	                 log2i(h) << 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);
+	BEGIN_RING(chan, sifm, NV04_SCALED_IMAGE_FROM_MEMORY_SURFACE, 1);
+	OUT_RING  (chan, swzsurf->handle);
+
+	for (cy = 0; cy < h; cy += sub_h) {
+	  for (cx = 0; cx < w; cx += sub_w) {
+	    BEGIN_RING(chan, swzsurf, NV04_SWIZZLED_SURFACE_OFFSET, 1);
+	    OUT_RELOCl(chan, dst_bo, dst->offset + nv04_swizzle_bits(cx, cy) *
+			     dst->block.size, 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, sub_h << 16 | sub_w);
+	    OUT_RING  (chan, 0);
+	    OUT_RING  (chan, sub_h << 16 | sub_w);
+	    OUT_RING  (chan, 1 << 20);
+	    OUT_RING  (chan, 1 << 20);
+
+	    BEGIN_RING(chan, sifm, NV04_SCALED_IMAGE_FROM_MEMORY_SIZE, 4);
+	    OUT_RING  (chan, sub_h << 16 | sub_w);
+	    OUT_RING  (chan, src->stride |
+	                     NV04_SCALED_IMAGE_FROM_MEMORY_FORMAT_ORIGIN_CENTER |
+	                     NV04_SCALED_IMAGE_FROM_MEMORY_FORMAT_FILTER_POINT_SAMPLE);
+	    OUT_RELOCl(chan, src_bo, src->offset + cy * src->stride +
+			     cx * src->block.size, NOUVEAU_BO_GART |
+			     NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
+	    OUT_RING  (chan, 0);
+	  }
+	}
+
+	return 0;
+}
+
+static int
+nv04_surface_copy_m2mf(struct nv04_surface_2d *ctx,
+		       struct pipe_surface *dst, int dx, int dy,
+		       struct pipe_surface *src, int sx, int sy, int w, int h)
+{
+	struct nouveau_channel *chan = ctx->nvws->channel;
+	struct nouveau_grobj *m2mf = ctx->m2mf;
+	struct nouveau_bo *src_bo = ctx->nvws->get_bo(ctx->buf(src));
+	struct nouveau_bo *dst_bo = ctx->nvws->get_bo(ctx->buf(dst));
+	unsigned dst_offset, src_offset;
+
+	dst_offset = dst->offset + (dy * dst->stride) + (dx * dst->block.size);
+	src_offset = src->offset + (sy * src->stride) + (sx * src->block.size);
+
+	WAIT_RING (chan, 3 + ((h / 2047) + 1) * 9);
+	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);
+	OUT_RELOCo(chan, dst_bo,
+		   NOUVEAU_BO_GART | NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
+
+	while (h) {
+		int count = (h > 2047) ? 2047 : h;
+
+		BEGIN_RING(chan, m2mf, NV04_MEMORY_TO_MEMORY_FORMAT_OFFSET_IN, 8);
+		OUT_RELOCl(chan, src_bo, src_offset,
+			   NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD);
+		OUT_RELOCl(chan, dst_bo, dst_offset,
+			   NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_WR);
+		OUT_RING  (chan, src->stride);
+		OUT_RING  (chan, dst->stride);
+		OUT_RING  (chan, w * src->block.size);
+		OUT_RING  (chan, count);
+		OUT_RING  (chan, 0x0101);
+		OUT_RING  (chan, 0);
+
+		h -= count;
+		src_offset += src->stride * count;
+		dst_offset += dst->stride * count;
+	}
+
+	return 0;
+}
+
+static int
+nv04_surface_copy_blit(struct nv04_surface_2d *ctx, struct pipe_surface *dst,
+		       int dx, int dy, struct pipe_surface *src, int sx, int sy,
+		       int w, int h)
+{
+	struct nouveau_channel *chan = ctx->nvws->channel;
+	struct nouveau_grobj *surf2d = ctx->surf2d;
+	struct nouveau_grobj *blit = ctx->blit;
+	struct nouveau_bo *src_bo = ctx->nvws->get_bo(ctx->buf(src));
+	struct nouveau_bo *dst_bo = ctx->nvws->get_bo(ctx->buf(dst));
+	int format;
+
+	format = nv04_surface_format(dst->format);
+	if (format < 0)
+		return 1;
+
+	WAIT_RING (chan, 12);
+	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);
+	BEGIN_RING(chan, surf2d, NV04_CONTEXT_SURFACES_2D_FORMAT, 4);
+	OUT_RING  (chan, format);
+	OUT_RING  (chan, (dst->stride << 16) | src->stride);
+	OUT_RELOCl(chan, src_bo, src->offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
+	OUT_RELOCl(chan, dst_bo, dst->offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
+
+	BEGIN_RING(chan, blit, 0x0300, 3);
+	OUT_RING  (chan, (sy << 16) | sx);
+	OUT_RING  (chan, (dy << 16) | dx);
+	OUT_RING  (chan, ( h << 16) |  w);
+
+	return 0;
+}
+
+static void
+nv04_surface_copy(struct nv04_surface_2d *ctx, struct pipe_surface *dst,
+		  int dx, int dy, struct pipe_surface *src, int sx, int sy,
+		  int w, int h)
+{
+	int src_linear = src->texture->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR;
+	int dst_linear = dst->texture->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR;
+
+	assert(src->format == dst->format);
+
+	/* Setup transfer to swizzle the texture to vram if needed */
+	/* FIXME/TODO: check proper limits of this operation */
+	if (src_linear ^ dst_linear) {
+		nv04_surface_copy_swizzle(ctx, dst, dx, dy, src, sx, sy, w, h);
+		return;
+	}
+
+	/* NV_CONTEXT_SURFACES_2D has buffer alignment restrictions, fallback
+	 * to NV_MEMORY_TO_MEMORY_FORMAT in this case.
+	 */
+	if ((src->offset & 63) || (dst->offset & 63)) {
+		nv04_surface_copy_m2mf(ctx, dst, dx, dy, src, sx, sy, w, h);
+		return;
+	}
+
+	nv04_surface_copy_blit(ctx, dst, dx, dy, src, sx, sy, w, h);
+}
+
+static void
+nv04_surface_fill(struct nv04_surface_2d *ctx, struct pipe_surface *dst,
+		  int dx, int dy, int w, int h, unsigned value)
+{
+	struct nouveau_channel *chan = ctx->nvws->channel;
+	struct nouveau_grobj *surf2d = ctx->surf2d;
+	struct nouveau_grobj *rect = ctx->rect;
+	struct nouveau_bo *dst_bo = ctx->nvws->get_bo(ctx->buf(dst));
+	int cs2d_format, gdirect_format;
+
+	cs2d_format = nv04_surface_format(dst->format);
+	assert(cs2d_format >= 0);
+
+	gdirect_format = nv04_surface_format(dst->format);
+	assert(gdirect_format >= 0);
+
+	WAIT_RING (chan, 16);
+	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);
+	BEGIN_RING(chan, surf2d, NV04_CONTEXT_SURFACES_2D_FORMAT, 4);
+	OUT_RING  (chan, cs2d_format);
+	OUT_RING  (chan, (dst->stride << 16) | dst->stride);
+	OUT_RELOCl(chan, dst_bo, dst->offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
+	OUT_RELOCl(chan, dst_bo, dst->offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
+
+	BEGIN_RING(chan, rect, NV04_GDI_RECTANGLE_TEXT_COLOR_FORMAT, 1);
+	OUT_RING  (chan, gdirect_format);
+	BEGIN_RING(chan, rect, NV04_GDI_RECTANGLE_TEXT_COLOR1_A, 1);
+	OUT_RING  (chan, value);
+	BEGIN_RING(chan, rect,
+		   NV04_GDI_RECTANGLE_TEXT_UNCLIPPED_RECTANGLE_POINT(0), 2);
+	OUT_RING  (chan, (dx << 16) | dy);
+	OUT_RING  (chan, ( w << 16) |  h);
+}
+
+void
+nv04_surface_2d_takedown(struct nv04_surface_2d **pctx)
+{
+	struct nv04_surface_2d *ctx;
+
+	if (!pctx || !*pctx)
+		return;
+	ctx = *pctx;
+	*pctx = NULL;
+
+	nouveau_notifier_free(&ctx->ntfy);
+	nouveau_grobj_free(&ctx->m2mf);
+	nouveau_grobj_free(&ctx->surf2d);
+	nouveau_grobj_free(&ctx->swzsurf);
+	nouveau_grobj_free(&ctx->rect);
+	nouveau_grobj_free(&ctx->blit);
+	nouveau_grobj_free(&ctx->sifm);
+
+	FREE(ctx);
+}
+
+struct nv04_surface_2d *
+nv04_surface_2d_init(struct nouveau_winsys *nvws)
+{
+	struct nv04_surface_2d *ctx = CALLOC_STRUCT(nv04_surface_2d);
+	struct nouveau_channel *chan = nvws->channel;
+	unsigned handle = 0x88000000, class;
+	int ret;
+
+	if (!ctx)
+		return NULL;
+
+	ret = nouveau_notifier_alloc(chan, handle++, 1, &ctx->ntfy);
+	if (ret) {
+		nv04_surface_2d_takedown(&ctx);
+		return NULL;
+	}
+
+	ret = nouveau_grobj_alloc(chan, handle++, 0x0039, &ctx->m2mf);
+	if (ret) {
+		nv04_surface_2d_takedown(&ctx);
+		return NULL;
+	}
+
+	BEGIN_RING(chan, ctx->m2mf, NV04_MEMORY_TO_MEMORY_FORMAT_DMA_NOTIFY, 1);
+	OUT_RING  (chan, ctx->ntfy->handle);
+
+	if (chan->device->chipset < 0x10)
+		class = NV04_CONTEXT_SURFACES_2D;
+	else
+		class = NV10_CONTEXT_SURFACES_2D;
+
+	ret = nouveau_grobj_alloc(chan, handle++, class, &ctx->surf2d);
+	if (ret) {
+		nv04_surface_2d_takedown(&ctx);
+		return NULL;
+	}
+
+	BEGIN_RING(chan, ctx->surf2d,
+			 NV04_CONTEXT_SURFACES_2D_DMA_IMAGE_SOURCE, 2);
+	OUT_RING  (chan, chan->vram->handle);
+	OUT_RING  (chan, chan->vram->handle);
+
+	if (chan->device->chipset < 0x10)
+		class = NV04_IMAGE_BLIT;
+	else
+		class = NV12_IMAGE_BLIT;
+
+	ret = nouveau_grobj_alloc(chan, handle++, class, &ctx->blit);
+	if (ret) {
+		nv04_surface_2d_takedown(&ctx);
+		return NULL;
+	}
+
+	BEGIN_RING(chan, ctx->blit, NV04_IMAGE_BLIT_DMA_NOTIFY, 1);
+	OUT_RING  (chan, ctx->ntfy->handle);
+	BEGIN_RING(chan, ctx->blit, NV04_IMAGE_BLIT_SURFACE, 1);
+	OUT_RING  (chan, ctx->surf2d->handle);
+	BEGIN_RING(chan, ctx->blit, NV04_IMAGE_BLIT_OPERATION, 1);
+	OUT_RING  (chan, NV04_IMAGE_BLIT_OPERATION_SRCCOPY);
+
+	ret = nouveau_grobj_alloc(chan, handle++, NV04_GDI_RECTANGLE_TEXT,
+				  &ctx->rect);
+	if (ret) {
+		nv04_surface_2d_takedown(&ctx);
+		return NULL;
+	}
+
+	BEGIN_RING(chan, ctx->rect, NV04_GDI_RECTANGLE_TEXT_DMA_NOTIFY, 1);
+	OUT_RING  (chan, ctx->ntfy->handle);
+	BEGIN_RING(chan, ctx->rect, NV04_GDI_RECTANGLE_TEXT_SURFACE, 1);
+	OUT_RING  (chan, ctx->ntfy->handle);
+	BEGIN_RING(chan, ctx->rect, NV04_GDI_RECTANGLE_TEXT_OPERATION, 1);
+	OUT_RING  (chan, NV04_GDI_RECTANGLE_TEXT_OPERATION_SRCCOPY);
+	BEGIN_RING(chan, ctx->rect,
+			 NV04_GDI_RECTANGLE_TEXT_MONOCHROME_FORMAT, 1);
+	OUT_RING  (chan, NV04_GDI_RECTANGLE_TEXT_MONOCHROME_FORMAT_LE);
+
+	switch (chan->device->chipset & 0xf0) {
+	case 0x00:
+	case 0x10:
+		class = NV04_SWIZZLED_SURFACE;
+		break;
+	case 0x20:
+		class = NV20_SWIZZLED_SURFACE;
+		break;
+	case 0x30:
+		class = NV30_SWIZZLED_SURFACE;
+		break;
+	case 0x40:
+	case 0x60:
+		class = NV40_SWIZZLED_SURFACE;
+		break;
+	default:
+		/* Famous last words: this really can't happen.. */
+		assert(0);
+		break;
+	}
+
+	ret = nouveau_grobj_alloc(chan, handle++, class, &ctx->swzsurf);
+	if (ret) {
+		nv04_surface_2d_takedown(&ctx);
+		return NULL;
+	}
+
+	if (chan->device->chipset < 0x10) {
+		class = NV04_SCALED_IMAGE_FROM_MEMORY;
+	} else
+	if (chan->device->chipset < 0x40) {
+		class = NV10_SCALED_IMAGE_FROM_MEMORY;
+	} else {
+		class = NV40_SCALED_IMAGE_FROM_MEMORY;
+	}
+
+	ret = nouveau_grobj_alloc(chan, handle++, class, &ctx->sifm);
+	if (ret) {
+		nv04_surface_2d_takedown(&ctx);
+		return NULL;
+	}
+
+	ctx->nvws = nvws;
+	ctx->copy = nv04_surface_copy;
+	ctx->fill = nv04_surface_fill;
+	return ctx;
+}
+
+
diff --git a/src/gallium/drivers/nv04/nv04_surface_2d.h b/src/gallium/drivers/nv04/nv04_surface_2d.h
new file mode 100644
index 0000000000..21b8f86960
--- /dev/null
+++ b/src/gallium/drivers/nv04/nv04_surface_2d.h
@@ -0,0 +1,29 @@
+#ifndef __NV04_SURFACE_2D_H__
+#define __NV04_SURFACE_2D_H__
+
+struct nv04_surface_2d {
+	struct nouveau_winsys *nvws;
+	struct nouveau_notifier *ntfy;
+	struct nouveau_grobj *surf2d;
+	struct nouveau_grobj *swzsurf;
+	struct nouveau_grobj *m2mf;
+	struct nouveau_grobj *rect;
+	struct nouveau_grobj *blit;
+	struct nouveau_grobj *sifm;
+
+	struct pipe_buffer *(*buf)(struct pipe_surface *);
+
+	void (*copy)(struct nv04_surface_2d *, struct pipe_surface *dst,
+		     int dx, int dy, struct pipe_surface *src, int sx, int sy,
+		     int w, int h);
+	void (*fill)(struct nv04_surface_2d *, struct pipe_surface *dst,
+		     int dx, int dy, int w, int h, unsigned value);
+};
+
+struct nv04_surface_2d *
+nv04_surface_2d_init(struct nouveau_winsys *nvws);
+
+void
+nv04_surface_2d_takedown(struct nv04_surface_2d **);
+
+#endif
diff --git a/src/gallium/drivers/nv10/nv10_screen.c b/src/gallium/drivers/nv10/nv10_screen.c
index 2f945a193c..f417b06c94 100644
--- a/src/gallium/drivers/nv10/nv10_screen.c
+++ b/src/gallium/drivers/nv10/nv10_screen.c
@@ -152,6 +152,14 @@ nv10_screen_destroy(struct pipe_screen *pscreen)
 	FREE(pscreen);
 }
 
+static struct pipe_buffer *
+nv10_surface_buffer(struct pipe_surface *surf)
+{
+	struct nv10_miptree *mt = (struct nv10_miptree *)surf->texture;
+
+	return mt->buffer;
+}
+
 struct pipe_screen *
 nv10_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws)
 {
@@ -164,6 +172,10 @@ nv10_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws)
 		return NULL;
 	screen->nvws = nvws;
 
+	/* 2D engine setup */
+	screen->eng2d = nv04_surface_2d_init(nvws);
+	screen->eng2d->buf = nv10_surface_buffer;
+
 	/* 3D object */
 	if (chipset>=0x20)
 		celsius_class=NV11TCL;
diff --git a/src/gallium/drivers/nv10/nv10_screen.h b/src/gallium/drivers/nv10/nv10_screen.h
index 3f8750a13f..60102a369a 100644
--- a/src/gallium/drivers/nv10/nv10_screen.h
+++ b/src/gallium/drivers/nv10/nv10_screen.h
@@ -2,6 +2,7 @@
 #define __NV10_SCREEN_H__
 
 #include "pipe/p_screen.h"
+#include "nv04/nv04_surface_2d.h"
 
 struct nv10_screen {
 	struct pipe_screen pipe;
@@ -9,6 +10,7 @@ struct nv10_screen {
 	struct nouveau_winsys *nvws;
 
 	/* HW graphics objects */
+	struct nv04_surface_2d *eng2d;
 	struct nouveau_grobj *celsius;
 	struct nouveau_notifier *sync;
 };
diff --git a/src/gallium/drivers/nv10/nv10_surface.c b/src/gallium/drivers/nv10/nv10_surface.c
index 78fd7b42da..1093dfd62e 100644
--- a/src/gallium/drivers/nv10/nv10_surface.c
+++ b/src/gallium/drivers/nv10/nv10_surface.c
@@ -39,10 +39,17 @@ nv10_surface_copy(struct pipe_context *pipe, boolean do_flip,
 		  unsigned width, unsigned height)
 {
 	struct nv10_context *nv10 = nv10_context(pipe);
-	struct nouveau_winsys *nvws = nv10->nvws;
+	struct nv04_surface_2d *eng2d = nv10->screen->eng2d;
 
-	nvws->surface_copy(nvws, dest, destx, desty, src, srcx, srcy,
-			   width, height);
+	if (do_flip) {
+		desty += height;
+		while (height--) {
+			eng2d->copy(eng2d, dest, destx, desty--, src,
+				    srcx, srcy++, width, 1);
+		}
+	}
+
+	eng2d->copy(eng2d, dest, destx, desty, src, srcx, srcy, width, height);
 }
 
 static void
@@ -51,9 +58,9 @@ nv10_surface_fill(struct pipe_context *pipe, struct pipe_surface *dest,
 		  unsigned height, unsigned value)
 {
 	struct nv10_context *nv10 = nv10_context(pipe);
-	struct nouveau_winsys *nvws = nv10->nvws;
+	struct nv04_surface_2d *eng2d = nv10->screen->eng2d;
 
-	nvws->surface_fill(nvws, dest, destx, desty, width, height, value);
+	eng2d->fill(eng2d, dest, destx, desty, width, height, value);
 }
 
 void
diff --git a/src/gallium/drivers/nv20/nv20_screen.c b/src/gallium/drivers/nv20/nv20_screen.c
index c9171fa178..5f2b7b4f71 100644
--- a/src/gallium/drivers/nv20/nv20_screen.c
+++ b/src/gallium/drivers/nv20/nv20_screen.c
@@ -152,6 +152,14 @@ nv20_screen_destroy(struct pipe_screen *pscreen)
 	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)
 {
@@ -164,6 +172,10 @@ nv20_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws)
 		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;
diff --git a/src/gallium/drivers/nv20/nv20_screen.h b/src/gallium/drivers/nv20/nv20_screen.h
index 8f2f2e341d..bf2f2c0d9f 100644
--- a/src/gallium/drivers/nv20/nv20_screen.h
+++ b/src/gallium/drivers/nv20/nv20_screen.h
@@ -2,6 +2,7 @@
 #define __NV20_SCREEN_H__
 
 #include "pipe/p_screen.h"
+#include "nv04/nv04_surface_2d.h"
 
 struct nv20_screen {
 	struct pipe_screen pipe;
@@ -9,6 +10,7 @@ struct nv20_screen {
 	struct nouveau_winsys *nvws;
 
 	/* HW graphics objects */
+	struct nv04_surface_2d *eng2d;
 	struct nouveau_grobj *kelvin;
 	struct nouveau_notifier *sync;
 };
diff --git a/src/gallium/drivers/nv20/nv20_surface.c b/src/gallium/drivers/nv20/nv20_surface.c
index 9b4c028eae..a79974ce5e 100644
--- a/src/gallium/drivers/nv20/nv20_surface.c
+++ b/src/gallium/drivers/nv20/nv20_surface.c
@@ -39,10 +39,17 @@ nv20_surface_copy(struct pipe_context *pipe, boolean do_flip,
 		  unsigned width, unsigned height)
 {
 	struct nv20_context *nv20 = nv20_context(pipe);
-	struct nouveau_winsys *nvws = nv20->nvws;
+	struct nv04_surface_2d *eng2d = nv20->screen->eng2d;
 
-	nvws->surface_copy(nvws, dest, destx, desty, src, srcx, srcy,
-			   width, height);
+	if (do_flip) {
+		desty += height;
+		while (height--) {
+			eng2d->copy(eng2d, dest, destx, desty--, src,
+				    srcx, srcy++, width, 1);
+		}
+	}
+
+	eng2d->copy(eng2d, dest, destx, desty, src, srcx, srcy, width, height);
 }
 
 static void
@@ -51,9 +58,9 @@ nv20_surface_fill(struct pipe_context *pipe, struct pipe_surface *dest,
 		  unsigned height, unsigned value)
 {
 	struct nv20_context *nv20 = nv20_context(pipe);
-	struct nouveau_winsys *nvws = nv20->nvws;
+	struct nv04_surface_2d *eng2d = nv20->screen->eng2d;
 
-	nvws->surface_fill(nvws, dest, destx, desty, width, height, value);
+	eng2d->fill(eng2d, dest, destx, desty, width, height, value);
 }
 
 void
diff --git a/src/gallium/drivers/nv30/nv30_screen.c b/src/gallium/drivers/nv30/nv30_screen.c
index 9738436dc4..2bc83f815b 100644
--- a/src/gallium/drivers/nv30/nv30_screen.c
+++ b/src/gallium/drivers/nv30/nv30_screen.c
@@ -220,6 +220,14 @@ nv30_screen_destroy(struct pipe_screen *pscreen)
 	FREE(pscreen);
 }
 
+static struct pipe_buffer *
+nv30_surface_buffer(struct pipe_surface *surf)
+{
+	struct nv30_miptree *mt = (struct nv30_miptree *)surf->texture;
+
+	return mt->buffer;
+}
+
 struct pipe_screen *
 nv30_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws)
 {
@@ -233,6 +241,10 @@ nv30_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws)
 		return NULL;
 	screen->nvws = nvws;
 
+	/* 2D engine setup */
+	screen->eng2d = nv04_surface_2d_init(nvws);
+	screen->eng2d->buf = nv30_surface_buffer;
+
 	/* 3D object */
 	switch (chipset & 0xf0) {
 	case 0x30:
diff --git a/src/gallium/drivers/nv30/nv30_screen.h b/src/gallium/drivers/nv30/nv30_screen.h
index b7ddc2a959..b11e470f94 100644
--- a/src/gallium/drivers/nv30/nv30_screen.h
+++ b/src/gallium/drivers/nv30/nv30_screen.h
@@ -2,6 +2,7 @@
 #define __NV30_SCREEN_H__
 
 #include "pipe/p_screen.h"
+#include "nv04/nv04_surface_2d.h"
 
 struct nv30_screen {
 	struct pipe_screen pipe;
@@ -11,6 +12,7 @@ struct nv30_screen {
 	unsigned cur_pctx;
 
 	/* HW graphics objects */
+	struct nv04_surface_2d *eng2d;
 	struct nouveau_grobj *rankine;
 	struct nouveau_notifier *sync;
 
diff --git a/src/gallium/drivers/nv30/nv30_surface.c b/src/gallium/drivers/nv30/nv30_surface.c
index 806131dcc9..b46b6123cf 100644
--- a/src/gallium/drivers/nv30/nv30_surface.c
+++ b/src/gallium/drivers/nv30/nv30_surface.c
@@ -30,7 +30,6 @@
 #include "pipe/p_defines.h"
 #include "pipe/internal/p_winsys_screen.h"
 #include "pipe/p_inlines.h"
-
 #include "util/u_tile.h"
 
 static void
@@ -40,22 +39,17 @@ nv30_surface_copy(struct pipe_context *pipe, boolean do_flip,
 		  unsigned width, unsigned height)
 {
 	struct nv30_context *nv30 = nv30_context(pipe);
-	struct nouveau_winsys *nvws = nv30->nvws;
+	struct nv04_surface_2d *eng2d = nv30->screen->eng2d;
 
 	if (do_flip) {
-		/*XXX: This dodgyness will do for now for correctness.  But,
-		 *     need to investigate whether the 2D engine is able to
-		 *     manage a flip (perhaps SIFM?), if not, use the 3D engine
-		 */
 		desty += height;
 		while (height--) {
-			nvws->surface_copy(nvws, dest, destx, desty--, src,
-					   srcx, srcy++, width, 1);
+			eng2d->copy(eng2d, dest, destx, desty--, src,
+				    srcx, srcy++, width, 1);
 		}
-	} else {
-		nvws->surface_copy(nvws, dest, destx, desty, src, srcx, srcy,
-				   width, height);
 	}
+
+	eng2d->copy(eng2d, dest, destx, desty, src, srcx, srcy, width, height);
 }
 
 static void
@@ -64,9 +58,9 @@ nv30_surface_fill(struct pipe_context *pipe, struct pipe_surface *dest,
 		  unsigned height, unsigned value)
 {
 	struct nv30_context *nv30 = nv30_context(pipe);
-	struct nouveau_winsys *nvws = nv30->nvws;
+	struct nv04_surface_2d *eng2d = nv30->screen->eng2d;
 
-	nvws->surface_fill(nvws, dest, destx, desty, width, height, value);
+	eng2d->fill(eng2d, dest, destx, desty, width, height, value);
 }
 
 void
diff --git a/src/gallium/drivers/nv40/nv40_screen.c b/src/gallium/drivers/nv40/nv40_screen.c
index 41d342d27d..a2b124d228 100644
--- a/src/gallium/drivers/nv40/nv40_screen.c
+++ b/src/gallium/drivers/nv40/nv40_screen.c
@@ -230,6 +230,14 @@ nv40_screen_destroy(struct pipe_screen *pscreen)
 	FREE(pscreen);
 }
 
+static struct pipe_buffer *
+nv40_surface_buffer(struct pipe_surface *surf)
+{
+	struct nv40_miptree *mt = (struct nv40_miptree *)surf->texture;
+
+	return mt->buffer;
+}
+
 struct pipe_screen *
 nv40_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws)
 {
@@ -243,6 +251,10 @@ nv40_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws)
 		return NULL;
 	screen->nvws = nvws;
 
+	/* 2D engine setup */
+	screen->eng2d = nv04_surface_2d_init(nvws);
+	screen->eng2d->buf = nv40_surface_buffer;
+
 	/* 3D object */
 	switch (chipset & 0xf0) {
 	case 0x40:
diff --git a/src/gallium/drivers/nv40/nv40_screen.h b/src/gallium/drivers/nv40/nv40_screen.h
index c04a1275a0..4500aa0e5c 100644
--- a/src/gallium/drivers/nv40/nv40_screen.h
+++ b/src/gallium/drivers/nv40/nv40_screen.h
@@ -2,6 +2,7 @@
 #define __NV40_SCREEN_H__
 
 #include "pipe/p_screen.h"
+#include "nv04/nv04_surface_2d.h"
 
 struct nv40_screen {
 	struct pipe_screen pipe;
@@ -11,6 +12,7 @@ struct nv40_screen {
 	unsigned cur_pctx;
 
 	/* HW graphics objects */
+	struct nv04_surface_2d *eng2d;
 	struct nouveau_grobj *curie;
 	struct nouveau_notifier *sync;
 
diff --git a/src/gallium/drivers/nv40/nv40_surface.c b/src/gallium/drivers/nv40/nv40_surface.c
index aa51d04051..68bbfce448 100644
--- a/src/gallium/drivers/nv40/nv40_surface.c
+++ b/src/gallium/drivers/nv40/nv40_surface.c
@@ -30,7 +30,6 @@
 #include "pipe/p_defines.h"
 #include "pipe/internal/p_winsys_screen.h"
 #include "pipe/p_inlines.h"
-
 #include "util/u_tile.h"
 
 static void
@@ -40,22 +39,17 @@ nv40_surface_copy(struct pipe_context *pipe, boolean do_flip,
 		  unsigned width, unsigned height)
 {
 	struct nv40_context *nv40 = nv40_context(pipe);
-	struct nouveau_winsys *nvws = nv40->nvws;
+	struct nv04_surface_2d *eng2d = nv40->screen->eng2d;
 
 	if (do_flip) {
-		/*XXX: This dodgyness will do for now for correctness.  But,
-		 *     need to investigate whether the 2D engine is able to
-		 *     manage a flip (perhaps SIFM?), if not, use the 3D engine
-		 */
 		desty += height;
 		while (height--) {
-			nvws->surface_copy(nvws, dest, destx, desty--, src,
-					   srcx, srcy++, width, 1);
+			eng2d->copy(eng2d, dest, destx, desty--, src,
+				    srcx, srcy++, width, 1);
 		}
-	} else {
-		nvws->surface_copy(nvws, dest, destx, desty, src, srcx, srcy,
-				   width, height);
 	}
+
+	eng2d->copy(eng2d, dest, destx, desty, src, srcx, srcy, width, height);
 }
 
 static void
@@ -64,9 +58,9 @@ nv40_surface_fill(struct pipe_context *pipe, struct pipe_surface *dest,
 		  unsigned height, unsigned value)
 {
 	struct nv40_context *nv40 = nv40_context(pipe);
-	struct nouveau_winsys *nvws = nv40->nvws;
+	struct nv04_surface_2d *eng2d = nv40->screen->eng2d;
 
-	nvws->surface_fill(nvws, dest, destx, desty, width, height, value);
+	eng2d->fill(eng2d, dest, destx, desty, width, height, value);
 }
 
 void