nouveau: nv04-nv40 texture transfer.

6 files changed, 256 insertions, 61 deletions

diff --git a/src/gallium/drivers/nv20/Makefile b/src/gallium/drivers/nv20/Makefile
index 93e34f8e92..1305f26c59 100644
--- a/src/gallium/drivers/nv20/Makefile
+++ b/src/gallium/drivers/nv20/Makefile
@@ -14,6 +14,7 @@ C_SOURCES = \
 	nv20_state.c \
 	nv20_state_emit.c \
 	nv20_surface.c \
+	nv20_transfer.c \
 	nv20_vbo.c
 #	nv20_vertprog.c
 
diff --git a/src/gallium/drivers/nv20/nv20_miptree.c b/src/gallium/drivers/nv20/nv20_miptree.c
index ef7e9c5428..2946240897 100644
--- a/src/gallium/drivers/nv20/nv20_miptree.c
+++ b/src/gallium/drivers/nv20/nv20_miptree.c
@@ -9,10 +9,14 @@ static void
 nv20_miptree_layout(struct nv20_miptree *nv20mt)
 {
 	struct pipe_texture *pt = &nv20mt->base;
-	boolean swizzled = FALSE;
 	uint width = pt->width[0], height = pt->height[0];
 	uint offset = 0;
 	int nr_faces, l, f;
+	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;
@@ -26,25 +30,31 @@ nv20_miptree_layout(struct nv20_miptree *nv20mt)
 		pt->nblocksx[l] = pf_get_nblocksx(&pt->block, width);
 		pt->nblocksy[l] = pf_get_nblocksy(&pt->block, height);
 
-		if (swizzled)
-			nv20mt->level[l].pitch = pt->nblocksx[l] * pt->block.size;
+		if (wide_pitch && (pt->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR))
+			nv20mt->level[l].pitch = align(pt->width[0] * pt->block.size, 64);
 		else
-			nv20mt->level[l].pitch = pt->nblocksx[0] * pt->block.size;
-		nv20mt->level[l].pitch = (nv20mt->level[l].pitch + 63) & ~63;
+			nv20mt->level[l].pitch = pt->width[l] * pt->block.size;
 
 		nv20mt->level[l].image_offset =
 			CALLOC(nr_faces, sizeof(unsigned));
 
 		width  = MAX2(1, width  >> 1);
 		height = MAX2(1, height >> 1);
-
 	}
 
 	for (f = 0; f < nr_faces; f++) {
-		for (l = 0; l <= pt->last_level; l++) {
+		for (l = 0; l < pt->last_level; l++) {
 			nv20mt->level[l].image_offset[f] = offset;
-			offset += nv20mt->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(nv20mt->level[l].pitch * pt->height[l], 64);
+			else
+				offset += nv20mt->level[l].pitch * pt->height[l];
 		}
+
+		nv20mt->level[l].image_offset[f] = offset;
+		offset += nv20mt->level[l].pitch * pt->height[l];
 	}
 
 	nv20mt->total_size = offset;
@@ -96,7 +106,8 @@ nv20_miptree_create(struct pipe_screen *screen, 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)
@@ -107,7 +118,11 @@ nv20_miptree_create(struct pipe_screen *screen, 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;
 		}
@@ -152,33 +167,33 @@ nv20_miptree_surface_get(struct pipe_screen *screen, struct pipe_texture *pt,
 			 unsigned flags)
 {
 	struct nv20_miptree *nv20mt = (struct nv20_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 = nv20mt->level[level].pitch;
-	ps->usage = flags;
-	ps->status = PIPE_SURFACE_STATUS_DEFINED;
-	ps->refcount = 1;
+	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 = nv20mt->level[level].pitch;
 
 	if (pt->target == PIPE_TEXTURE_CUBE) {
-		ps->offset = nv20mt->level[level].image_offset[face];
+		ns->base.offset = nv20mt->level[level].image_offset[face];
 	} else
 	if (pt->target == PIPE_TEXTURE_3D) {
-		ps->offset = nv20mt->level[level].image_offset[zslice];
+		ns->base.offset = nv20mt->level[level].image_offset[zslice];
 	} else {
-		ps->offset = nv20mt->level[level].image_offset[0];
+		ns->base.offset = nv20mt->level[level].image_offset[0];
 	}
 
-	return ps;
+	return &ns->base;
 }
 
 static void
diff --git a/src/gallium/drivers/nv20/nv20_screen.c b/src/gallium/drivers/nv20/nv20_screen.c
index 5f2b7b4f71..7760ae27c0 100644
--- a/src/gallium/drivers/nv20/nv20_screen.c
+++ b/src/gallium/drivers/nv20/nv20_screen.c
@@ -116,30 +116,6 @@ nv20_screen_is_format_supported(struct pipe_screen *screen,
 	return FALSE;
 }
 
-static void *
-nv20_surface_map(struct pipe_screen *screen, struct pipe_surface *surface,
-		 unsigned flags )
-{
-	struct pipe_winsys *ws = screen->winsys;
-	void *map;
-	struct nv20_miptree *nv20mt = (struct nv20_miptree *)surface->texture;
-
-	map = ws->buffer_map(ws, nv20mt->buffer, flags);
-	if (!map)
-		return NULL;
-
-	return map + surface->offset;
-}
-
-static void
-nv20_surface_unmap(struct pipe_screen *screen, struct pipe_surface *surface)
-{
-	struct pipe_winsys *ws = screen->winsys;
-	struct nv20_miptree *nv20mt = (struct nv20_miptree *)surface->texture;
-
-	ws->buffer_unmap(ws, nv20mt->buffer);
-}
-
 static void
 nv20_screen_destroy(struct pipe_screen *pscreen)
 {
@@ -211,10 +187,8 @@ nv20_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws)
 
 	screen->pipe.is_format_supported = nv20_screen_is_format_supported;
 
-	screen->pipe.surface_map = nv20_surface_map;
-	screen->pipe.surface_unmap = nv20_surface_unmap;
-
 	nv20_screen_init_miptree_functions(&screen->pipe);
+	nv20_screen_init_transfer_functions(&screen->pipe);
 	u_simple_screen_init(&screen->pipe);
 
 	return &screen->pipe;
diff --git a/src/gallium/drivers/nv20/nv20_screen.h b/src/gallium/drivers/nv20/nv20_screen.h
index bf2f2c0d9f..d9fce2bced 100644
--- a/src/gallium/drivers/nv20/nv20_screen.h
+++ b/src/gallium/drivers/nv20/nv20_screen.h
@@ -21,4 +21,8 @@ nv20_screen(struct pipe_screen *screen)
 	return (struct nv20_screen *)screen;
 }
 
+
+void
+nv20_screen_init_transfer_functions(struct pipe_screen *pscreen);
+
 #endif
diff --git a/src/gallium/drivers/nv20/nv20_state_emit.c b/src/gallium/drivers/nv20/nv20_state_emit.c
index 0f4df9ca31..4042f46d05 100644
--- a/src/gallium/drivers/nv20/nv20_state_emit.c
+++ b/src/gallium/drivers/nv20/nv20_state_emit.c
@@ -109,7 +109,7 @@ static void nv20_state_emit_scissor(struct nv20_context* nv20)
 static void nv20_state_emit_framebuffer(struct nv20_context* nv20)
 {
 	struct pipe_framebuffer_state* fb = nv20->framebuffer;
-	struct pipe_surface *rt, *zeta = NULL;
+	struct nv04_surface *rt, *zeta = NULL;
 	uint32_t rt_format, w, h;
 	int colour_format = 0, zeta_format = 0;
 	struct nv20_miptree *nv20mt = 0;
@@ -117,7 +117,7 @@ static void nv20_state_emit_framebuffer(struct nv20_context* nv20)
 	w = fb->cbufs[0]->width;
 	h = fb->cbufs[0]->height;
 	colour_format = fb->cbufs[0]->format;
-	rt = fb->cbufs[0];
+	rt = (struct nv04_surface *)fb->cbufs[0];
 
 	if (fb->zsbuf) {
 		if (colour_format) {
@@ -129,7 +129,7 @@ static void nv20_state_emit_framebuffer(struct nv20_context* nv20)
 		}
 
 		zeta_format = fb->zsbuf->format;
-		zeta = fb->zsbuf;
+		zeta = (struct nv04_surface *)fb->zsbuf;
 	}
 
 	rt_format = NV20TCL_RT_FORMAT_TYPE_LINEAR | 0x20;
@@ -148,18 +148,18 @@ static void nv20_state_emit_framebuffer(struct nv20_context* nv20)
 
 	if (zeta) {
 		BEGIN_RING(kelvin, NV20TCL_RT_PITCH, 1);
-		OUT_RING  (rt->stride | (zeta->stride << 16));
+		OUT_RING  (rt->pitch | (zeta->pitch << 16));
 	} else {
 		BEGIN_RING(kelvin, NV20TCL_RT_PITCH, 1);
-		OUT_RING  (rt->stride | (rt->stride << 16));
+		OUT_RING  (rt->pitch | (rt->pitch << 16));
 	}
 
-	nv20mt = (struct nv20_miptree *)rt->texture;
+	nv20mt = (struct nv20_miptree *)rt->base.texture;
 	nv20->rt[0] = nv20mt->buffer;
 
 	if (zeta_format)
 	{
-		nv20mt = (struct nv20_miptree *)zeta->texture;
+		nv20mt = (struct nv20_miptree *)zeta->base.texture;
 		nv20->zeta = nv20mt->buffer;
 	}
 
diff --git a/src/gallium/drivers/nv20/nv20_transfer.c b/src/gallium/drivers/nv20/nv20_transfer.c
new file mode 100644
index 0000000000..19de09486d
--- /dev/null
+++ b/src/gallium/drivers/nv20/nv20_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 "nv20_context.h"
+#include "nv20_screen.h"
+#include "nv20_state.h"
+
+struct nv20_transfer {
+	struct pipe_transfer base;
+	struct pipe_surface *surface;
+	bool direct;
+};
+
+static unsigned nv20_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
+nv20_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 *
+nv20_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 nv20_miptree *mt = (struct nv20_miptree *)pt;
+	struct nv20_transfer *tx;
+	struct pipe_texture tx_tex_template, *tx_tex;
+
+	tx = CALLOC_STRUCT(nv20_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,
+	                                               nv20_usage_tx_to_buf(usage));
+		return &tx->base;
+	}
+
+	tx->direct = false;
+
+	nv20_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,
+	                                       nv20_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 nv20_screen *nvscreen = nv20_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
+nv20_transfer_del(struct pipe_screen *pscreen, struct pipe_transfer **pptx)
+{
+	struct pipe_transfer *ptx = *pptx;
+	struct nv20_transfer *tx = (struct nv20_transfer *)ptx;
+
+	if (!tx->direct && ptx->usage != PIPE_TRANSFER_READ) {
+		struct nv20_screen *nvscreen = nv20_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 *
+nv20_transfer_map(struct pipe_screen *pscreen, struct pipe_transfer *ptx)
+{
+	struct nv20_transfer *tx = (struct nv20_transfer *)ptx;
+	struct nv04_surface *ns = (struct nv04_surface *)tx->surface;
+	struct nv20_miptree *mt = (struct nv20_miptree *)tx->surface->texture;
+	void *map = pipe_buffer_map(pscreen, mt->buffer,
+	                            nv20_usage_tx_to_buf(ptx->usage));
+
+	return map + ns->base.offset +
+	       ptx->y * ns->pitch + ptx->x * ptx->block.size;
+}
+
+static void
+nv20_transfer_unmap(struct pipe_screen *pscreen, struct pipe_transfer *ptx)
+{
+	struct nv20_transfer *tx = (struct nv20_transfer *)ptx;
+	struct nv20_miptree *mt = (struct nv20_miptree *)tx->surface->texture;
+
+	pipe_buffer_unmap(pscreen, mt->buffer);
+}
+
+void
+nv20_screen_init_transfer_functions(struct pipe_screen *pscreen)
+{
+	pscreen->get_tex_transfer = nv20_transfer_new;
+	pscreen->tex_transfer_release = nv20_transfer_del;
+	pscreen->transfer_map = nv20_transfer_map;
+	pscreen->transfer_unmap = nv20_transfer_unmap;
+}