freedreno: gmem bypass

The GPU (at least a3xx, but I think also a2xx) can render directly to
memory, bypassing tiling.  Although it can't do this if blend, depth,
and a few other features of the pipeline are enabled.  This direct
memory mode can be faster for some sorts of operations, such as simple
blits.  In particular, this significantly speeds up XA by avoiding to
pull the entire dest pixmap into GMEM, render tiles, and write it all
back out again.  This should also speed up resource copy-region and
blit.

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

1 files changed, 20 insertions, 0 deletions

diff --git a/src/gallium/drivers/freedreno/freedreno_context.h b/src/gallium/drivers/freedreno/freedreno_context.h
index a6133c0d8c3..54759314e26 100644
--- a/src/gallium/drivers/freedreno/freedreno_context.h
+++ b/src/gallium/drivers/freedreno/freedreno_context.h
@@ -117,6 +117,23 @@ struct fd_context {
 
 	bool needs_flush;
 
+	/* To decide whether to render to system memory, keep track of the
+	 * number of draws, and whether any of them require multisample,
+	 * depth_test (or depth write), stencil_test, blending, and
+	 * color_logic_Op (since those functions are disabled when by-
+	 * passing GMEM.
+	 */
+	enum {
+		FD_GMEM_CLEARS_DEPTH_STENCIL = 0x01,
+		FD_GMEM_DEPTH_ENABLED        = 0x02,
+		FD_GMEM_STENCIL_ENABLED      = 0x04,
+
+		FD_GMEM_MSAA_ENABLED         = 0x08,
+		FD_GMEM_BLEND_ENABLED        = 0x10,
+		FD_GMEM_LOGICOP_ENABLED      = 0x20,
+	} gmem_reason;
+	unsigned num_draws;
+
 	struct fd_ringbuffer *ring;
 	struct fd_ringmarker *draw_start, *draw_end;
 
@@ -186,6 +203,9 @@ struct fd_context {
 	void (*emit_tile_gmem2mem)(struct fd_context *ctx, uint32_t xoff, uint32_t yoff,
 			uint32_t bin_w, uint32_t bin_h);
 
+	/* optional, for GMEM bypass: */
+	void (*emit_sysmem_prep)(struct fd_context *ctx);
+
 	/* draw: */
 	void (*draw)(struct fd_context *pctx, const struct pipe_draw_info *info);
 	void (*clear)(struct fd_context *ctx, unsigned buffers,