freedreno/a5xx: texture tiling

Overall a nice 5-10% gain for most games.  And more for things like
glmark2 texture benchmark.

There are some rough edges.  In particular, the hardware seems to only
support tiling or component swap.  (Ie. from hw PoV, ARGB/ABGR/RGBA/
BGRA are all the same format but with different component swap.)  For
tiled formats, only ARGB is possible.  This isn't a big problem for
*sampling* since we also have swizzle state there (and since
util_format_compose_swizzles() already takes into account the component
order, we didn't use COLOR_SWAP for sampling).  But it is a problem if
you try to render to a tiled BGRA (for example) surface.

The next patch introduces a workaround for blitter, so we can generate
tiled textures in ABGR/RGBA/BGRA, but that doesn't help the render-
target case.  To handle that, I think we'd need to keep track that the
tiled format is different from the linear format, which seems like it
would get extra fun with sampler views/etc.

So for now, disabled by default, enable with FD_MESA_DEBUG=ttile.  In
practice it works fine for all the games I've tried, but makes piglit
grumpy.

Signed-off-by: Rob Clark <[email protected]>

1 files changed, 78 insertions, 8 deletions

diff --git a/src/gallium/drivers/freedreno/freedreno_resource.c b/src/gallium/drivers/freedreno/freedreno_resource.c
index eaaf2e8b229..7e6de8c057f 100644
--- a/src/gallium/drivers/freedreno/freedreno_resource.c
+++ b/src/gallium/drivers/freedreno/freedreno_resource.c
@@ -290,6 +290,7 @@ fd_alloc_staging(struct fd_context *ctx, struct fd_resource *rsc,
 	tmpl.depth0  = box->depth;
 	tmpl.array_size = 1;
 	tmpl.last_level = 0;
+	tmpl.bind |= PIPE_BIND_LINEAR;
 
 	struct pipe_resource *pstaging =
 		pctx->screen->resource_create(pctx->screen, &tmpl);
@@ -300,7 +301,7 @@ fd_alloc_staging(struct fd_context *ctx, struct fd_resource *rsc,
 }
 
 static void
-fd_blit_staging(struct fd_context *ctx, struct fd_transfer *trans)
+fd_blit_from_staging(struct fd_context *ctx, struct fd_transfer *trans)
 {
 	struct pipe_resource *dst = trans->base.resource;
 	struct pipe_blit_info blit = {0};
@@ -317,7 +318,26 @@ fd_blit_staging(struct fd_context *ctx, struct fd_transfer *trans)
 	blit.filter = PIPE_TEX_FILTER_NEAREST;
 
 	do_blit(ctx, &blit, false);
-	pipe_resource_reference(&trans->staging_prsc, NULL);
+}
+
+static void
+fd_blit_to_staging(struct fd_context *ctx, struct fd_transfer *trans)
+{
+	struct pipe_resource *src = trans->base.resource;
+	struct pipe_blit_info blit = {0};
+
+	blit.src.resource = src;
+	blit.src.format   = src->format;
+	blit.src.level    = trans->base.level;
+	blit.src.box      = trans->base.box;
+	blit.dst.resource = trans->staging_prsc;
+	blit.dst.format   = trans->staging_prsc->format;
+	blit.dst.level    = 0;
+	blit.dst.box      = trans->staging_box;
+	blit.mask = util_format_get_mask(trans->staging_prsc->format);
+	blit.filter = PIPE_TEX_FILTER_NEAREST;
+
+	do_blit(ctx, &blit, false);
 }
 
 static unsigned
@@ -396,8 +416,11 @@ fd_resource_transfer_unmap(struct pipe_context *pctx,
 	struct fd_resource *rsc = fd_resource(ptrans->resource);
 	struct fd_transfer *trans = fd_transfer(ptrans);
 
-	if (trans->staging_prsc)
-		fd_blit_staging(ctx, trans);
+	if (trans->staging_prsc) {
+		if (ptrans->usage & PIPE_TRANSFER_WRITE)
+			fd_blit_from_staging(ctx, trans);
+		pipe_resource_reference(&trans->staging_prsc, NULL);
+	}
 
 	if (!(ptrans->usage & PIPE_TRANSFER_UNSYNCHRONIZED)) {
 		fd_bo_cpu_fini(rsc->bo);
@@ -447,6 +470,44 @@ fd_resource_transfer_map(struct pipe_context *pctx,
 	ptrans->stride = util_format_get_nblocksx(format, slice->pitch) * rsc->cpp;
 	ptrans->layer_stride = rsc->layer_first ? rsc->layer_size : slice->size0;
 
+	/* we always need a staging texture for tiled buffers:
+	 *
+	 * TODO we might sometimes want to *also* shadow the resource to avoid
+	 * splitting a batch.. for ex, mid-frame texture uploads to a tiled
+	 * texture.
+	 */
+	if (rsc->tile_mode) {
+		struct fd_resource *staging_rsc;
+
+		staging_rsc = fd_alloc_staging(ctx, rsc, level, box);
+		if (staging_rsc) {
+			// TODO for PIPE_TRANSFER_READ, need to do untiling blit..
+			trans->staging_prsc = &staging_rsc->base;
+			trans->base.stride = util_format_get_nblocksx(format,
+				staging_rsc->slices[0].pitch) * staging_rsc->cpp;
+			trans->base.layer_stride = staging_rsc->layer_first ?
+				staging_rsc->layer_size : staging_rsc->slices[0].size0;
+			trans->staging_box = *box;
+			trans->staging_box.x = 0;
+			trans->staging_box.y = 0;
+			trans->staging_box.z = 0;
+
+			if (usage & PIPE_TRANSFER_READ) {
+				fd_blit_to_staging(ctx, trans);
+				fd_bo_cpu_prep(rsc->bo, ctx->pipe, DRM_FREEDRENO_PREP_READ);
+			}
+
+			buf = fd_bo_map(staging_rsc->bo);
+			offset = 0;
+
+			*pptrans = ptrans;
+
+			ctx->stats.staging_uploads++;
+
+			return buf;
+		}
+	}
+
 	if (ctx->in_shadow && !(usage & PIPE_TRANSFER_READ))
 		usage |= PIPE_TRANSFER_UNSYNCHRONIZED;
 
@@ -616,9 +677,6 @@ setup_slices(struct fd_resource *rsc, uint32_t alignment, enum pipe_format forma
 	 */
 	uint32_t layers_in_level = rsc->layer_first ? 1 : prsc->array_size;
 
-	if (is_a5xx(screen) && (rsc->base.target >= PIPE_TEXTURE_2D))
-		height = align(height, screen->gmem_alignh);
-
 	for (level = 0; level <= prsc->last_level; level++) {
 		struct fd_resource_slice *slice = fd_resource_slice(rsc, level);
 		uint32_t blocks;
@@ -686,7 +744,7 @@ fd_setup_slices(struct fd_resource *rsc)
 	alignment = slice_alignment(rsc->base.target);
 
 	struct fd_screen *screen = fd_screen(rsc->base.screen);
-	if (is_a4xx(screen) || is_a5xx(screen)) {
+	if (is_a4xx(screen)) {
 		switch (rsc->base.target) {
 		case PIPE_TEXTURE_3D:
 			rsc->layer_first = false;
@@ -759,6 +817,18 @@ fd_resource_create(struct pipe_screen *pscreen,
 
 	*prsc = *tmpl;
 
+#define LINEAR \
+	(PIPE_BIND_SCANOUT | \
+	 PIPE_BIND_LINEAR  | \
+	 PIPE_BIND_DISPLAY_TARGET)
+
+	if (screen->tile_mode &&
+			(tmpl->target != PIPE_BUFFER) &&
+			(tmpl->bind & PIPE_BIND_SAMPLER_VIEW) &&
+			!(tmpl->bind & LINEAR)) {
+		rsc->tile_mode = screen->tile_mode(tmpl);
+	}
+
 	pipe_reference_init(&prsc->reference, 1);
 
 	prsc->screen = pscreen;