nouveau: nv04-nv40 texture transfer.

7 files changed, 244 insertions, 128 deletions

diff --git a/src/gallium/drivers/nv30/Makefile b/src/gallium/drivers/nv30/Makefile
index 4c29e2eab3..364c80d8f3 100644
--- a/src/gallium/drivers/nv30/Makefile
+++ b/src/gallium/drivers/nv30/Makefile
@@ -22,6 +22,7 @@ C_SOURCES = \
 	nv30_state_viewport.c \
 	nv30_state_zsa.c \
 	nv30_surface.c \
+	nv30_transfer.c \
 	nv30_vbo.c \
 	nv30_vertprog.c
 
diff --git a/src/gallium/drivers/nv30/nv30_miptree.c b/src/gallium/drivers/nv30/nv30_miptree.c
index b11ed8c24e..ec0a8b8438 100644
--- a/src/gallium/drivers/nv30/nv30_miptree.c
+++ b/src/gallium/drivers/nv30/nv30_miptree.c
@@ -69,6 +69,8 @@ nv30_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *pt)
 {
 	struct pipe_winsys *ws = pscreen->winsys;
 	struct nv30_miptree *mt;
+	unsigned buf_usage = PIPE_BUFFER_USAGE_PIXEL |
+	                     NOUVEAU_BUFFER_USAGE_TEXTURE;
 
 	mt = MALLOC(sizeof(struct nv30_miptree));
 	if (!mt)
@@ -76,8 +78,6 @@ nv30_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *pt)
 	mt->base = *pt;
 	mt->base.refcount = 1;
 	mt->base.screen = pscreen;
-	mt->shadow_tex = NULL;
-	mt->shadow_surface = NULL;
 
 	/* Swizzled textures must be POT */
 	if (pt->width[0] & (pt->width[0] - 1) ||
@@ -107,11 +107,12 @@ nv30_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *pt)
 		}
 	}
 
+	if (pt->tex_usage & PIPE_TEXTURE_USAGE_DYNAMIC)
+		buf_usage |= PIPE_BUFFER_USAGE_CPU_READ_WRITE;
+
 	nv30_miptree_layout(mt);
 
-	mt->buffer = ws->buffer_create(ws, 256,
-				       PIPE_BUFFER_USAGE_PIXEL |
-				       NOUVEAU_BUFFER_USAGE_TEXTURE,
+	mt->buffer = ws->buffer_create(ws, 256, buf_usage,
 				       mt->total_size);
 	if (!mt->buffer) {
 		FREE(mt);
@@ -163,12 +164,6 @@ nv30_miptree_release(struct pipe_screen *pscreen, struct pipe_texture **ppt)
 			FREE(mt->level[l].image_offset);
 	}
 
-	if (mt->shadow_tex) {
-		if (mt->shadow_surface)
-			pscreen->tex_surface_release(pscreen, &mt->shadow_surface);
-		nv30_miptree_release(pscreen, &mt->shadow_tex);
-	}
-
 	FREE(mt);
 }
 
@@ -178,36 +173,33 @@ nv30_miptree_surface_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 			 unsigned flags)
 {
 	struct nv30_miptree *nv30mt = (struct nv30_miptree *)pt;
-	struct pipe_surface *ps;
+	struct nv04_surface *ns;
 
-	ps = CALLOC_STRUCT(pipe_surface);
-	if (!ps)
+	ns = CALLOC_STRUCT(nv04_surface);
+	if (!ns)
 		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 = nv30mt->level[level].pitch;
-	ps->usage = flags;
-	ps->status = PIPE_SURFACE_STATUS_DEFINED;
-	ps->refcount = 1;
-	ps->face = face;
-	ps->level = level;
-	ps->zslice = zslice;
+	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.usage = flags;
+	ns->base.status = PIPE_SURFACE_STATUS_DEFINED;
+	ns->base.refcount = 1;
+	ns->base.face = face;
+	ns->base.level = level;
+	ns->base.zslice = zslice;
+	ns->pitch = nv30mt->level[level].pitch;
 
 	if (pt->target == PIPE_TEXTURE_CUBE) {
-		ps->offset = nv30mt->level[level].image_offset[face];
+		ns->base.offset = nv30mt->level[level].image_offset[face];
 	} else
 	if (pt->target == PIPE_TEXTURE_3D) {
-		ps->offset = nv30mt->level[level].image_offset[zslice];
+		ns->base.offset = nv30mt->level[level].image_offset[zslice];
 	} else {
-		ps->offset = nv30mt->level[level].image_offset[0];
+		ns->base.offset = nv30mt->level[level].image_offset[0];
 	}
 
-	return ps;
+	return &ns->base;
 }
 
 static void
diff --git a/src/gallium/drivers/nv30/nv30_screen.c b/src/gallium/drivers/nv30/nv30_screen.c
index c97a73f0b1..d395c5e1b7 100644
--- a/src/gallium/drivers/nv30/nv30_screen.c
+++ b/src/gallium/drivers/nv30/nv30_screen.c
@@ -135,82 +135,6 @@ nv30_surface_buffer(struct pipe_surface *surf)
 	return mt->buffer;
 }
 
-static void *
-nv30_surface_map(struct pipe_screen *screen, struct pipe_surface *surface,
-		 unsigned flags )
-{
-	struct pipe_winsys	*ws = screen->winsys;
-	struct pipe_surface	*surface_to_map;
-	void			*map;
-
-	if (!(surface->texture->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR)) {
-		struct nv30_miptree *mt = (struct nv30_miptree *)surface->texture;
-
-		if (!mt->shadow_tex) {
-			unsigned old_tex_usage = surface->texture->tex_usage;
-			surface->texture->tex_usage = NOUVEAU_TEXTURE_USAGE_LINEAR |
-			                              PIPE_TEXTURE_USAGE_DYNAMIC;
-			mt->shadow_tex = screen->texture_create(screen, surface->texture);
-			surface->texture->tex_usage = old_tex_usage;
-
-			assert(mt->shadow_tex->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR);
-		}
-
-		mt->shadow_surface = screen->get_tex_surface
-		(
-			screen, mt->shadow_tex,
-			surface->face, surface->level, surface->zslice,
-			surface->usage
-		);
-
-		surface_to_map = mt->shadow_surface;
-	}
-	else
-		surface_to_map = surface;
-
-	assert(surface_to_map);
-
-	map = ws->buffer_map(ws, nv30_surface_buffer(surface_to_map), flags);
-	if (!map)
-		return NULL;
-
-	return map + surface_to_map->offset;
-}
-
-static void
-nv30_surface_unmap(struct pipe_screen *screen, struct pipe_surface *surface)
-{
-	struct pipe_winsys	*ws = screen->winsys;
-	struct pipe_surface	*surface_to_unmap;
-
-	/* TODO: Copy from shadow just before push buffer is flushed instead.
-	         There are probably some programs that map/unmap excessively
-	         before rendering. */
-	if (!(surface->texture->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR)) {
-		struct nv30_miptree *mt = (struct nv30_miptree *)surface->texture;
-
-		assert(mt->shadow_tex);
-
-		surface_to_unmap = mt->shadow_surface;
-	}
-	else
-		surface_to_unmap = surface;
-
-	assert(surface_to_unmap);
-
-	ws->buffer_unmap(ws, nv30_surface_buffer(surface_to_unmap));
-
-	if (surface_to_unmap != surface) {
-		struct nv30_screen *nvscreen = nv30_screen(screen);
-
-		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);
-	}
-}
-
 static void
 nv30_screen_destroy(struct pipe_screen *pscreen)
 {
@@ -391,10 +315,8 @@ nv30_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws)
 
 	screen->pipe.is_format_supported = nv30_screen_surface_format_supported;
 
-	screen->pipe.surface_map = nv30_surface_map;
-	screen->pipe.surface_unmap = nv30_surface_unmap;
-
 	nv30_screen_init_miptree_functions(&screen->pipe);
+	nv30_screen_init_transfer_functions(&screen->pipe);
 	u_simple_screen_init(&screen->pipe);
 
 	return &screen->pipe;
diff --git a/src/gallium/drivers/nv30/nv30_screen.h b/src/gallium/drivers/nv30/nv30_screen.h
index b11e470f94..8e36883975 100644
--- a/src/gallium/drivers/nv30/nv30_screen.h
+++ b/src/gallium/drivers/nv30/nv30_screen.h
@@ -34,4 +34,7 @@ nv30_screen(struct pipe_screen *screen)
 	return (struct nv30_screen *)screen;
 }
 
+void
+nv30_screen_init_transfer_functions(struct pipe_screen *pscreen);
+
 #endif
diff --git a/src/gallium/drivers/nv30/nv30_state.h b/src/gallium/drivers/nv30/nv30_state.h
index 2023278e37..e6f23bf166 100644
--- a/src/gallium/drivers/nv30/nv30_state.h
+++ b/src/gallium/drivers/nv30/nv30_state.h
@@ -76,9 +76,6 @@ struct nv30_miptree {
 	struct pipe_buffer *buffer;
 	uint total_size;
 
-	struct pipe_texture *shadow_tex;
-	struct pipe_surface *shadow_surface;
-
 	struct {
 		uint pitch;
 		uint *image_offset;
diff --git a/src/gallium/drivers/nv30/nv30_state_fb.c b/src/gallium/drivers/nv30/nv30_state_fb.c
index 77368cb205..ee0326011c 100644
--- a/src/gallium/drivers/nv30/nv30_state_fb.c
+++ b/src/gallium/drivers/nv30/nv30_state_fb.c
@@ -5,7 +5,7 @@ static boolean
 nv30_state_framebuffer_validate(struct nv30_context *nv30)
 {
 	struct pipe_framebuffer_state *fb = &nv30->framebuffer;
-	struct pipe_surface *rt[2], *zeta = NULL;
+	struct nv04_surface *rt[2], *zeta = NULL;
 	uint32_t rt_enable, rt_format;
 	int i, colour_format = 0, zeta_format = 0;
 	struct nouveau_stateobj *so = so_new(64, 10);
@@ -21,7 +21,7 @@ nv30_state_framebuffer_validate(struct nv30_context *nv30)
 		} else {
 			colour_format = fb->cbufs[i]->format;
 			rt_enable |= (NV34TCL_RT_ENABLE_COLOR0 << i);
-			rt[i] = fb->cbufs[i];
+			rt[i] = (struct nv04_surface *)fb->cbufs[i];
 		}
 	}
 
@@ -30,13 +30,13 @@ nv30_state_framebuffer_validate(struct nv30_context *nv30)
 
 	if (fb->zsbuf) {
 		zeta_format = fb->zsbuf->format;
-		zeta = fb->zsbuf;
+		zeta = (struct nv04_surface *)fb->zsbuf;
 	}
 
-	if (!(rt[0]->texture->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR)) {
+	if (!(rt[0]->base.texture->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR)) {
 		assert(!(fb->width & (fb->width - 1)) && !(fb->height & (fb->height - 1)));
 		for (i = 1; i < fb->nr_cbufs; i++)
-			assert(!(rt[i]->texture->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR));
+			assert(!(rt[i]->base.texture->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR));
 
 		/* FIXME: NV34TCL_RT_FORMAT_LOG2_[WIDTH/HEIGHT] */
 		rt_format = NV34TCL_RT_FORMAT_TYPE_SWIZZLED |
@@ -71,44 +71,44 @@ nv30_state_framebuffer_validate(struct nv30_context *nv30)
 	}
 
 	if (rt_enable & NV34TCL_RT_ENABLE_COLOR0) {
-		uint32_t pitch = rt[0]->stride;
+		uint32_t pitch = rt[0]->pitch;
 		if (zeta) {
-			pitch |= (zeta->stride << 16);
+			pitch |= (zeta->pitch << 16);
 		} else {
 			pitch |= (pitch << 16);
 		}
 
-		nv30mt = (struct nv30_miptree *)rt[0]->texture;
+		nv30mt = (struct nv30_miptree *)rt[0]->base.texture;
 		so_method(so, nv30->screen->rankine, NV34TCL_DMA_COLOR0, 1);
 		so_reloc (so, nv30mt->buffer, 0, rt_flags | NOUVEAU_BO_OR,
 			  nv30->nvws->channel->vram->handle,
 			  nv30->nvws->channel->gart->handle);
 		so_method(so, nv30->screen->rankine, NV34TCL_COLOR0_PITCH, 2);
 		so_data  (so, pitch);
-		so_reloc (so, nv30mt->buffer, rt[0]->offset, rt_flags |
+		so_reloc (so, nv30mt->buffer, rt[0]->base.offset, rt_flags |
 			  NOUVEAU_BO_LOW, 0, 0);
 	}
 
 	if (rt_enable & NV34TCL_RT_ENABLE_COLOR1) {
-		nv30mt = (struct nv30_miptree *)rt[1]->texture;
+		nv30mt = (struct nv30_miptree *)rt[1]->base.texture;
 		so_method(so, nv30->screen->rankine, NV34TCL_DMA_COLOR1, 1);
 		so_reloc (so, nv30mt->buffer, 0, rt_flags | NOUVEAU_BO_OR,
 			  nv30->nvws->channel->vram->handle,
 			  nv30->nvws->channel->gart->handle);
 		so_method(so, nv30->screen->rankine, NV34TCL_COLOR1_OFFSET, 2);
-		so_reloc (so, nv30mt->buffer, rt[1]->offset, rt_flags |
+		so_reloc (so, nv30mt->buffer, rt[1]->base.offset, rt_flags |
 			  NOUVEAU_BO_LOW, 0, 0);
-		so_data  (so, rt[1]->stride);
+		so_data  (so, rt[1]->pitch);
 	}
 
 	if (zeta_format) {
-		nv30mt = (struct nv30_miptree *)zeta->texture;
+		nv30mt = (struct nv30_miptree *)zeta->base.texture;
 		so_method(so, nv30->screen->rankine, NV34TCL_DMA_ZETA, 1);
 		so_reloc (so, nv30mt->buffer, 0, rt_flags | NOUVEAU_BO_OR,
 			  nv30->nvws->channel->vram->handle,
 			  nv30->nvws->channel->gart->handle);
 		so_method(so, nv30->screen->rankine, NV34TCL_ZETA_OFFSET, 1);
-		so_reloc (so, nv30mt->buffer, zeta->offset, rt_flags |
+		so_reloc (so, nv30mt->buffer, zeta->base.offset, rt_flags |
 			  NOUVEAU_BO_LOW, 0, 0);
 		/* TODO: allocate LMA depth buffer */
 	}
diff --git a/src/gallium/drivers/nv30/nv30_transfer.c b/src/gallium/drivers/nv30/nv30_transfer.c
new file mode 100644
index 0000000000..df4dc4b1f6
--- /dev/null
+++ b/src/gallium/drivers/nv30/nv30_transfer.c
@@ -0,0 +1,201 @@
+#include <pipe/p_state.h>
+#include <pipe/p_defines.h>
+#include <pipe/p_inlines.h>
+#include <util/u_memory.h>
+#include <nouveau/nouveau_winsys.h>
+#include "nv30_context.h"
+#include "nv30_screen.h"
+#include "nv30_state.h"
+
+struct nv30_transfer {
+	struct pipe_transfer base;
+	struct pipe_surface *surface;
+	bool direct;
+};
+
+static unsigned nv30_usage_tx_to_buf(unsigned tx_usage)
+{
+	switch (tx_usage) {
+		case PIPE_TRANSFER_READ:
+			return PIPE_BUFFER_USAGE_CPU_READ;
+		case PIPE_TRANSFER_WRITE:
+			return PIPE_BUFFER_USAGE_CPU_WRITE;
+		case PIPE_TRANSFER_READ_WRITE:
+			return PIPE_BUFFER_USAGE_CPU_READ_WRITE;
+		default:
+			assert(0);
+	}
+
+	return -1;
+}
+
+static void
+nv30_compatible_transfer_tex(struct pipe_texture *pt, unsigned level,
+                             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->last_level = 0;
+	template->compressed = pt->compressed;
+	template->nr_samples = pt->nr_samples;
+
+	template->tex_usage = PIPE_TEXTURE_USAGE_DYNAMIC |
+	                      NOUVEAU_TEXTURE_USAGE_LINEAR;
+}
+
+static struct pipe_transfer *
+nv30_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
+		  unsigned face, unsigned level, unsigned zslice,
+		  enum pipe_transfer_usage usage,
+		  unsigned x, unsigned y, unsigned w, unsigned h)
+{
+	struct nv30_miptree *mt = (struct nv30_miptree *)pt;
+	struct nv30_transfer *tx;
+	struct pipe_texture tx_tex_template, *tx_tex;
+
+	tx = CALLOC_STRUCT(nv30_transfer);
+	if (!tx)
+		return NULL;
+
+	tx->base.refcount = 1;
+	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;
+	tx->base.level = level;
+	tx->base.zslice = zslice;
+
+	/* Direct access to texture */
+	if ((pt->tex_usage & PIPE_TEXTURE_USAGE_DYNAMIC ||
+	     debug_get_bool_option("NOUVEAU_NO_TRANSFER", TRUE/*XXX:FALSE*/)) &&
+	    pt->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR)
+	{
+		tx->direct = true;
+		tx->surface = pscreen->get_tex_surface(pscreen, pt,
+	                                               face, level, zslice,
+	                                               nv30_usage_tx_to_buf(usage));
+		return &tx->base;
+	}
+
+	tx->direct = false;
+
+	nv30_compatible_transfer_tex(pt, level, &tx_tex_template);
+
+	tx_tex = pscreen->texture_create(pscreen, &tx_tex_template);
+	if (!tx_tex)
+	{
+		FREE(tx);
+		return NULL;
+	}
+
+	tx->surface = pscreen->get_tex_surface(pscreen, tx_tex,
+	                                       face, level, zslice,
+	                                       nv30_usage_tx_to_buf(usage));
+
+	pipe_texture_reference(&tx_tex, NULL);
+
+	if (!tx->surface)
+	{
+		pipe_surface_reference(&tx->surface, NULL);
+		FREE(tx);
+		return NULL;
+	}
+
+	if (usage != PIPE_TRANSFER_WRITE) {
+		struct nv30_screen *nvscreen = nv30_screen(pscreen);
+		struct pipe_surface *src;
+
+		src = pscreen->get_tex_surface(pscreen, pt,
+	                                       face, level, zslice,
+	                                       PIPE_BUFFER_USAGE_GPU_READ);
+
+		/* TODO: Check if SIFM can deal with x,y,w,h when swizzling */
+		/* TODO: Check if SIFM can un-swizzle */
+		nvscreen->eng2d->copy(nvscreen->eng2d,
+		                      tx->surface, 0, 0,
+		                      src, 0, 0,
+		                      src->width, src->height);
+
+		pipe_surface_reference(&src, NULL);
+	}
+
+	return &tx->base;
+}
+
+static void
+nv30_transfer_del(struct pipe_screen *pscreen, struct pipe_transfer **pptx)
+{
+	struct pipe_transfer *ptx = *pptx;
+	struct nv30_transfer *tx = (struct nv30_transfer *)ptx;
+
+	if (!tx->direct && ptx->usage != PIPE_TRANSFER_READ) {
+		struct nv30_screen *nvscreen = nv30_screen(pscreen);
+		struct pipe_surface *dst;
+
+		dst = pscreen->get_tex_surface(pscreen, ptx->texture,
+	                                       ptx->face, ptx->level, ptx->zslice,
+	                                       PIPE_BUFFER_USAGE_GPU_WRITE);
+
+		/* TODO: Check if SIFM can deal with x,y,w,h when swizzling */
+		nvscreen->eng2d->copy(nvscreen->eng2d,
+		                      dst, 0, 0,
+		                      tx->surface, 0, 0,
+		                      dst->width, dst->height);
+
+		pipe_surface_reference(&dst, NULL);
+	}
+
+	*pptx = NULL;
+	if (--ptx->refcount)
+		return;
+
+	pipe_surface_reference(&tx->surface, NULL);
+	pipe_texture_reference(&ptx->texture, NULL);
+	FREE(ptx);
+}
+
+static void *
+nv30_transfer_map(struct pipe_screen *pscreen, struct pipe_transfer *ptx)
+{
+	struct nv30_transfer *tx = (struct nv30_transfer *)ptx;
+	struct nv04_surface *ns = (struct nv04_surface *)tx->surface;
+	struct nv30_miptree *mt = (struct nv30_miptree *)tx->surface->texture;
+	void *map = pipe_buffer_map(pscreen, mt->buffer,
+	                            nv30_usage_tx_to_buf(ptx->usage));
+
+	return map + ns->base.offset +
+	       ptx->y * ns->pitch + ptx->x * ptx->block.size;
+}
+
+static void
+nv30_transfer_unmap(struct pipe_screen *pscreen, struct pipe_transfer *ptx)
+{
+	struct nv30_transfer *tx = (struct nv30_transfer *)ptx;
+	struct nv30_miptree *mt = (struct nv30_miptree *)tx->surface->texture;
+
+	pipe_buffer_unmap(pscreen, mt->buffer);
+}
+
+void
+nv30_screen_init_transfer_functions(struct pipe_screen *pscreen)
+{
+	pscreen->get_tex_transfer = nv30_transfer_new;
+	pscreen->tex_transfer_release = nv30_transfer_del;
+	pscreen->transfer_map = nv30_transfer_map;
+	pscreen->transfer_unmap = nv30_transfer_unmap;
+}