Merge branch 'master' into gallium-texture-transfer

Conflicts:

	src/mesa/state_tracker/st_cb_accum.c
	src/mesa/state_tracker/st_cb_drawpixels.c

5 files changed, 84 insertions, 35 deletions

diff --git a/src/gallium/drivers/nv40/nv40_miptree.c b/src/gallium/drivers/nv40/nv40_miptree.c
index b1fba11d2fe..e38b1e7f5ca 100644
--- a/src/gallium/drivers/nv40/nv40_miptree.c
+++ b/src/gallium/drivers/nv40/nv40_miptree.c
@@ -10,7 +10,12 @@ nv40_miptree_layout(struct nv40_miptree *mt)
 	struct pipe_texture *pt = &mt->base;
 	uint width = pt->width[0], height = pt->height[0], depth = pt->depth[0];
 	uint offset = 0;
-	int nr_faces, l, f, pitch;
+	int nr_faces, l, f;
+	uint wide_pitch = pt->tex_usage & (PIPE_TEXTURE_USAGE_SAMPLER |
+		                           PIPE_TEXTURE_USAGE_DEPTH_STENCIL |
+		                           PIPE_TEXTURE_USAGE_RENDER_TARGET |
+		                           PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
+		                           PIPE_TEXTURE_USAGE_PRIMARY);
 
 	if (pt->target == PIPE_TEXTURE_CUBE) {
 		nr_faces = 6;
@@ -21,7 +26,6 @@ nv40_miptree_layout(struct nv40_miptree *mt)
 		nr_faces = 1;
 	}
 
-	pitch = pt->width[0];
 	for (l = 0; l <= pt->last_level; l++) {
 		pt->width[l] = width;
 		pt->height[l] = height;
@@ -29,11 +33,11 @@ nv40_miptree_layout(struct nv40_miptree *mt)
 		pt->nblocksx[l] = pf_get_nblocksx(&pt->block, width);
 		pt->nblocksy[l] = pf_get_nblocksy(&pt->block, height);
 
-		if (!(pt->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR))
-			pitch = pt->nblocksx[l];
-		pitch = align(pitch, 64);
+		if (wide_pitch && (pt->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR))
+			mt->level[l].pitch = align(pt->width[0] * pt->block.size, 64);
+		else
+			mt->level[l].pitch = pt->width[l] * pt->block.size;
 
-		mt->level[l].pitch = pitch * pt->block.size;
 		mt->level[l].image_offset =
 			CALLOC(nr_faces, sizeof(unsigned));
 
@@ -43,10 +47,18 @@ nv40_miptree_layout(struct nv40_miptree *mt)
 	}
 
 	for (f = 0; f < nr_faces; f++) {
-		for (l = 0; l <= pt->last_level; l++) {
+		for (l = 0; l < pt->last_level; l++) {
 			mt->level[l].image_offset[f] = offset;
-			offset += mt->level[l].pitch * pt->height[l];
+
+			if (!(pt->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR) &&
+			    pt->width[l + 1] > 1 && pt->height[l + 1] > 1)
+				offset += align(mt->level[l].pitch * pt->height[l], 64);
+			else
+				offset += mt->level[l].pitch * pt->height[l];
 		}
+
+		mt->level[l].image_offset[f] = offset;
+		offset += mt->level[l].pitch * pt->height[l];
 	}
 
 	mt->total_size = offset;
@@ -75,7 +87,8 @@ nv40_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *pt)
 		mt->base.tex_usage |= NOUVEAU_TEXTURE_USAGE_LINEAR;
 	else
 	if (pt->tex_usage & (PIPE_TEXTURE_USAGE_PRIMARY |
-	                     PIPE_TEXTURE_USAGE_DISPLAY_TARGET))
+	                     PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
+	                     PIPE_TEXTURE_USAGE_DEPTH_STENCIL))
 		mt->base.tex_usage |= NOUVEAU_TEXTURE_USAGE_LINEAR;
 	else
 	if (pt->tex_usage & PIPE_TEXTURE_USAGE_DYNAMIC)
@@ -86,7 +99,11 @@ nv40_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *pt)
 		case PIPE_FORMAT_A8R8G8B8_UNORM:
 		case PIPE_FORMAT_X8R8G8B8_UNORM:
 		case PIPE_FORMAT_R16_SNORM:
+		{
+			if (debug_get_bool_option("NOUVEAU_NO_SWIZZLE", FALSE))
+				mt->base.tex_usage |= NOUVEAU_TEXTURE_USAGE_LINEAR;
 			break;
+		}
 		default:
 			mt->base.tex_usage |= NOUVEAU_TEXTURE_USAGE_LINEAR;
 		}
@@ -106,6 +123,31 @@ nv40_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *pt)
 	return &mt->base;
 }
 
+static struct pipe_texture *
+nv40_miptree_blanket(struct pipe_screen *pscreen, const struct pipe_texture *pt,
+		     const unsigned *stride, struct pipe_buffer *pb)
+{
+	struct nv40_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(nv40_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 void
 nv40_miptree_release(struct pipe_screen *pscreen, struct pipe_texture **ppt)
 {
@@ -124,8 +166,8 @@ nv40_miptree_release(struct pipe_screen *pscreen, struct pipe_texture **ppt)
 	}
 
 	if (mt->shadow_tex) {
-		assert(mt->shadow_surface);
-		pscreen->tex_surface_release(pscreen, &mt->shadow_surface);
+		if (mt->shadow_surface)
+			pscreen->tex_surface_release(pscreen, &mt->shadow_surface);
 		nv40_miptree_release(pscreen, &mt->shadow_tex);
 	}
 
@@ -188,6 +230,7 @@ void
 nv40_screen_init_miptree_functions(struct pipe_screen *pscreen)
 {
 	pscreen->texture_create = nv40_miptree_create;
+	pscreen->texture_blanket = nv40_miptree_blanket;
 	pscreen->texture_release = nv40_miptree_release;
 	pscreen->get_tex_surface = nv40_miptree_surface_new;
 	pscreen->tex_surface_release = nv40_miptree_surface_del;
diff --git a/src/gallium/drivers/nv40/nv40_screen.c b/src/gallium/drivers/nv40/nv40_screen.c
index 41d342d27d2..2372bc84410 100644
--- a/src/gallium/drivers/nv40/nv40_screen.c
+++ b/src/gallium/drivers/nv40/nv40_screen.c
@@ -136,6 +136,14 @@ nv40_screen_surface_format_supported(struct pipe_screen *pscreen,
 	return FALSE;
 }
 
+static struct pipe_buffer *
+nv40_surface_buffer(struct pipe_surface *surf)
+{
+	struct nv40_miptree *mt = (struct nv40_miptree *)surf->texture;
+
+	return mt->buffer;
+}
+
 static void *
 nv40_surface_map(struct pipe_screen *screen, struct pipe_surface *surface,
 		 unsigned flags )
@@ -143,7 +151,6 @@ nv40_surface_map(struct pipe_screen *screen, struct pipe_surface *surface,
 	struct pipe_winsys	*ws = screen->winsys;
 	struct pipe_surface	*surface_to_map;
 	void			*map;
-        struct nv40_miptree *mt;
 
 	if (!(surface->texture->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR)) {
 		struct nv40_miptree *mt = (struct nv40_miptree *)surface->texture;
@@ -171,8 +178,7 @@ nv40_surface_map(struct pipe_screen *screen, struct pipe_surface *surface,
 		surface_to_map = surface;
 
 	assert(surface_to_map);
-        mt = (struct nv40_miptree *)surface_to_map->texture;
-	map = ws->buffer_map(ws, mt->buffer, flags);
+	map = ws->buffer_map(ws, nv40_surface_buffer(surface_to_map), flags);
 	if (!map)
 		return NULL;
 
@@ -184,7 +190,6 @@ nv40_surface_unmap(struct pipe_screen *screen, struct pipe_surface *surface)
 {
 	struct pipe_winsys	*ws = screen->winsys;
 	struct pipe_surface	*surface_to_unmap;
-        struct nv40_miptree *mt;
 
 	/* TODO: Copy from shadow just before push buffer is flushed instead.
 	         There are probably some programs that map/unmap excessively
@@ -201,16 +206,16 @@ nv40_surface_unmap(struct pipe_screen *screen, struct pipe_surface *surface)
 
 	assert(surface_to_unmap);
 
-        mt = (struct nv40_miptree *)surface_to_unmap->texture;
-	ws->buffer_unmap(ws, mt->buffer);
+	ws->buffer_unmap(ws, nv40_surface_buffer(surface_to_unmap));
 
 	if (surface_to_unmap != surface) {
 		struct nv40_screen *nvscreen = nv40_screen(screen);
 
-		nvscreen->nvws->surface_copy(nvscreen->nvws,
-		                             surface, 0, 0,
-		                             surface_to_unmap, 0, 0,
-		                             surface->width, surface->height);
+		nvscreen->eng2d->copy(nvscreen->eng2d, surface, 0, 0,
+		                      surface_to_unmap, 0, 0,
+		                      surface->width, surface->height);
+
+		screen->tex_surface_release(screen, &surface_to_unmap);
 	}
 }
 
@@ -243,6 +248,10 @@ 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 c04a1275a00..4500aa0e5cc 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 aa51d040519..c4a5fb20d97 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 +39,18 @@ 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);
+		return;
 	}
+
+	eng2d->copy(eng2d, dest, destx, desty, src, srcx, srcy, width, height);
 }
 
 static void
@@ -64,9 +59,9 @@ 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
diff --git a/src/gallium/drivers/nv40/nv40_vertprog.c b/src/gallium/drivers/nv40/nv40_vertprog.c
index 1392fe956f7..0862386638c 100644
--- a/src/gallium/drivers/nv40/nv40_vertprog.c
+++ b/src/gallium/drivers/nv40/nv40_vertprog.c
@@ -784,7 +784,7 @@ nv40_vertprog_translate(struct nv40_context *nv40,
 
 			imm = &parse.FullToken.FullImmediate;
 			assert(imm->Immediate.DataType == TGSI_IMM_FLOAT32);
-//			assert(imm->Immediate.Size == 4);
+			assert(imm->Immediate.NrTokens == 4 + 1);
 			vpc->imm[vpc->nr_imm++] =
 				constant(vpc, -1,
 					 imm->u.ImmediateFloat32[0].Float,