nv50: fix textures with block size != cpp

First, using width * block size as pitch is evidently
wrong if a block contains more than 1 texel.

For tiled textures, since a block occupies a contiguous
area of memory, y addressing in m2mf has to be done by
block index, not the y coordinate itself.

This should fix compressed textures.

2 files changed, 48 insertions, 37 deletions

diff --git a/src/gallium/drivers/nv50/nv50_miptree.c b/src/gallium/drivers/nv50/nv50_miptree.c
index 93479a0314..229a59cb74 100644
--- a/src/gallium/drivers/nv50/nv50_miptree.c
+++ b/src/gallium/drivers/nv50/nv50_miptree.c
@@ -26,6 +26,16 @@
 
 #include "nv50_context.h"
 
+static INLINE uint32_t
+get_tile_mode(unsigned ny)
+{
+	if (ny > 32) return 4;
+	if (ny > 16) return 3;
+	if (ny >  8) return 2;
+	if (ny >  4) return 1;
+	return 0;
+}
+
 static struct pipe_texture *
 nv50_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *tmp)
 {
@@ -34,7 +44,7 @@ nv50_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *tmp)
 	struct pipe_texture *pt = &mt->base.base;
 	unsigned width = tmp->width[0], height = tmp->height[0];
 	unsigned depth = tmp->depth[0];
-	uint32_t tile_mode, tile_flags, tile_h;
+	uint32_t tile_flags;
 	int ret, i, l;
 
 	*pt = *tmp;
@@ -57,13 +67,6 @@ nv50_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *tmp)
 		break;
 	}
 
-	if      (pt->height[0] > 32) tile_mode = 4;
-	else if (pt->height[0] > 16) tile_mode = 3;
-	else if (pt->height[0] >  8) tile_mode = 2;
-	else if (pt->height[0] >  4) tile_mode = 1;
-	else                         tile_mode = 0;
-	tile_h = 1 << (tile_mode + 2);
-
 	switch (pt->target) {
 	case PIPE_TEXTURE_3D:
 		mt->image_nr = pt->depth[0];
@@ -86,28 +89,22 @@ nv50_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *tmp)
 		pt->nblocksy[l] = pf_get_nblocksy(&pt->block, height);
 
 		lvl->image_offset = CALLOC(mt->image_nr, sizeof(int));
-		lvl->pitch = align(pt->width[l] * pt->block.size, 64);
-		lvl->tile_mode = tile_mode;
+		lvl->pitch = align(pt->nblocksx[l] * pt->block.size, 64);
+		lvl->tile_mode = get_tile_mode(pt->nblocksy[l]);
 
 		width = MAX2(1, width >> 1);
 		height = MAX2(1, height >> 1);
 		depth = MAX2(1, depth >> 1);
-
-		if (tile_mode && height <= (tile_h >> 1)) {
-			tile_mode--;
-			tile_h >>= 1;
-		}
 	}
 
 	for (i = 0; i < mt->image_nr; i++) {
 		for (l = 0; l <= pt->last_level; l++) {
 			struct nv50_miptree_level *lvl = &mt->level[l];
 			int size;
-			tile_h = 1 << (lvl->tile_mode + 2);
+			unsigned tile_ny = 1 << (lvl->tile_mode + 2);
 
-			size  = align(pt->width[l], 8) * pt->block.size;
-			size  = align(size, 64);
-			size *= align(pt->height[l], tile_h);
+			size  = align(pt->nblocksx[l] * pt->block.size, 64);
+			size *= align(pt->nblocksy[l], tile_ny);
 
 			lvl->image_offset[i] = mt->total_size;
 
diff --git a/src/gallium/drivers/nv50/nv50_transfer.c b/src/gallium/drivers/nv50/nv50_transfer.c
index f1eb672336..9c008090b8 100644
--- a/src/gallium/drivers/nv50/nv50_transfer.c
+++ b/src/gallium/drivers/nv50/nv50_transfer.c
@@ -124,7 +124,7 @@ nv50_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 	struct nv50_miptree *mt = nv50_miptree(pt);
 	struct nv50_miptree_level *lvl = &mt->level[level];
 	struct nv50_transfer *tx;
-	unsigned image = 0;
+	unsigned nx, ny, image = 0;
 	int ret;
 
 	if (pt->target == PIPE_TEXTURE_CUBE)
@@ -142,9 +142,16 @@ nv50_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 	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 = (w * pt->block.size);
+	if (!pt->nblocksx[level]) {
+		tx->base.nblocksx = pf_get_nblocksx(&pt->block,
+						    pt->width[level]);
+		tx->base.nblocksy = pf_get_nblocksy(&pt->block,
+						    pt->height[level]);
+	} else {
+		tx->base.nblocksx = pt->nblocksx[level];
+		tx->base.nblocksy = pt->nblocksy[level];
+	}
+	tx->base.stride = tx->base.nblocksx * pt->block.size;
 	tx->base.usage = usage;
 
 	tx->level_pitch = lvl->pitch;
@@ -152,24 +159,28 @@ nv50_transfer_new(struct pipe_screen *pscreen, struct pipe_texture *pt,
 	tx->level_height = mt->base.base.height[level];
 	tx->level_offset = lvl->image_offset[image];
 	tx->level_tiling = lvl->tile_mode;
-	tx->level_x = x;
-	tx->level_y = y;
+	tx->level_x = pf_get_nblocksx(&tx->base.block, x);
+	tx->level_y = pf_get_nblocksy(&tx->base.block, y);
 	ret = nouveau_bo_new(dev, NOUVEAU_BO_GART | NOUVEAU_BO_MAP, 0,
-			     w * pt->block.size * h, &tx->bo);
+			     tx->base.nblocksy * tx->base.stride, &tx->bo);
 	if (ret) {
 		FREE(tx);
 		return NULL;
 	}
 
 	if (usage & PIPE_TRANSFER_READ) {
+		nx = pf_get_nblocksx(&tx->base.block, tx->base.width);
+		ny = pf_get_nblocksy(&tx->base.block, tx->base.height);
+
 		nv50_transfer_rect_m2mf(pscreen, mt->base.bo, tx->level_offset,
 					tx->level_pitch, tx->level_tiling,
 					x, y,
-					tx->level_width, tx->level_height,
-					tx->bo, 0, tx->base.stride,
-					tx->bo->tile_mode, 0, 0,
-					tx->base.width, tx->base.height,
-					tx->base.block.size, w, h,
+					tx->base.nblocksx, tx->base.nblocksy,
+					tx->bo, 0,
+					tx->base.stride, tx->bo->tile_mode,
+					0, 0,
+					tx->base.nblocksx, tx->base.nblocksy,
+					tx->base.block.size, nx, ny,
 					NOUVEAU_BO_VRAM | NOUVEAU_BO_GART,
 					NOUVEAU_BO_GART);
 	}
@@ -183,17 +194,20 @@ nv50_transfer_del(struct pipe_transfer *ptx)
 	struct nv50_transfer *tx = (struct nv50_transfer *)ptx;
 	struct nv50_miptree *mt = nv50_miptree(ptx->texture);
 
+	unsigned nx = pf_get_nblocksx(&tx->base.block, tx->base.width);
+	unsigned ny = pf_get_nblocksy(&tx->base.block, tx->base.height);
+
 	if (ptx->usage & PIPE_TRANSFER_WRITE) {
 		struct pipe_screen *pscreen = ptx->texture->screen;
-		nv50_transfer_rect_m2mf(pscreen, tx->bo, 0, tx->base.stride,
-					tx->bo->tile_mode, 0, 0,
-					tx->base.width, tx->base.height,
+		nv50_transfer_rect_m2mf(pscreen, tx->bo, 0,
+					tx->base.stride, tx->bo->tile_mode,
+					0, 0,
+					tx->base.nblocksx, tx->base.nblocksy,
 					mt->base.bo, tx->level_offset,
 					tx->level_pitch, tx->level_tiling,
 					tx->level_x, tx->level_y,
-					tx->level_width, tx->level_height,
-					tx->base.block.size, tx->base.width,
-					tx->base.height,
+					tx->base.nblocksx, tx->base.nblocksy,
+					tx->base.block.size, nx, ny,
 					NOUVEAU_BO_GART, NOUVEAU_BO_VRAM |
 					NOUVEAU_BO_GART);
 	}