drm/vc4: Add exec flags to allow forcing a specific X/Y tile walk order.

This is useful to allow GL to provide defined results for overlapping
glBlitFramebuffer, which X11 in turn uses to accelerate uncomposited
window movement without first blitting to a temporary.  x11perf
-copywinwin100 goes from 1850/sec to 4850/sec.

v2: Default to the same behavior as before when the flags aren't
    passed. (suggested by Boris)

Signed-off-by: Eric Anholt <[email protected]>
Link: https://patchwork.freedesktop.org/patch/msgid/[email protected]
Reviewed-by: Boris Brezillon <[email protected]>

drivers/gpu/drm/vc4/vc4_drv.c

@@ -99,6 +99,7 @@ static int vc4_get_param_ioctl(struct drm_device *dev, void *data,
 	case DRM_VC4_PARAM_SUPPORTS_BRANCHES:
 	case DRM_VC4_PARAM_SUPPORTS_ETC1:
 	case DRM_VC4_PARAM_SUPPORTS_THREADED_FS:
+	case DRM_VC4_PARAM_SUPPORTS_FIXED_RCL_ORDER:
 		args->value = true;
 		break;
 	default:

drivers/gpu/drm/vc4/vc4_gem.c

@@ -1007,7 +1007,10 @@ vc4_submit_cl_ioctl(struct drm_device *dev, void *data,
 	struct ww_acquire_ctx acquire_ctx;
 	int ret = 0;
 
-	if ((args->flags & ~VC4_SUBMIT_CL_USE_CLEAR_COLOR) != 0) {
+	if ((args->flags & ~(VC4_SUBMIT_CL_USE_CLEAR_COLOR |
+			     VC4_SUBMIT_CL_FIXED_RCL_ORDER |
+			     VC4_SUBMIT_CL_RCL_ORDER_INCREASING_X |
+			     VC4_SUBMIT_CL_RCL_ORDER_INCREASING_Y)) != 0) {
 		DRM_DEBUG("Unknown flags: 0x%02x\n", args->flags);
 		return -EINVAL;
 	}

drivers/gpu/drm/vc4/vc4_render_cl.c

@@ -261,8 +261,17 @@ static int vc4_create_rcl_bo(struct drm_device *dev, struct vc4_exec_info *exec,
 	uint8_t max_y_tile = args->max_y_tile;
 	uint8_t xtiles = max_x_tile - min_x_tile + 1;
 	uint8_t ytiles = max_y_tile - min_y_tile + 1;
-	uint8_t x, y;
+	uint8_t xi, yi;
 	uint32_t size, loop_body_size;
+	bool positive_x = true;
+	bool positive_y = true;
+
+	if (args->flags & VC4_SUBMIT_CL_FIXED_RCL_ORDER) {
+		if (!(args->flags & VC4_SUBMIT_CL_RCL_ORDER_INCREASING_X))
+			positive_x = false;
+		if (!(args->flags & VC4_SUBMIT_CL_RCL_ORDER_INCREASING_Y))
+			positive_y = false;
+	}
 
 	size = VC4_PACKET_TILE_RENDERING_MODE_CONFIG_SIZE;
 	loop_body_size = VC4_PACKET_TILE_COORDINATES_SIZE;
@@ -354,10 +363,12 @@ static int vc4_create_rcl_bo(struct drm_device *dev, struct vc4_exec_info *exec,
 	rcl_u16(setup, args->height);
 	rcl_u16(setup, args->color_write.bits);
 
-	for (y = min_y_tile; y <= max_y_tile; y++) {
-		for (x = min_x_tile; x <= max_x_tile; x++) {
-			bool first = (x == min_x_tile && y == min_y_tile);
-			bool last = (x == max_x_tile && y == max_y_tile);
+	for (yi = 0; yi < ytiles; yi++) {
+		int y = positive_y ? min_y_tile + yi : max_y_tile - yi;
+		for (xi = 0; xi < xtiles; xi++) {
+			int x = positive_x ? min_x_tile + xi : max_x_tile - xi;
+			bool first = (xi == 0 && yi == 0);
+			bool last = (xi == xtiles - 1 && yi == ytiles - 1);
 
 			emit_tile(exec, setup, x, y, first, last);
 		}

include/uapi/drm/vc4_drm.h

@@ -155,6 +155,16 @@ struct drm_vc4_submit_cl {
 	__u32 pad:24;
 
 #define VC4_SUBMIT_CL_USE_CLEAR_COLOR			(1 << 0)
+/* By default, the kernel gets to choose the order that the tiles are
+ * rendered in.  If this is set, then the tiles will be rendered in a
+ * raster order, with the right-to-left vs left-to-right and
+ * top-to-bottom vs bottom-to-top dictated by
+ * VC4_SUBMIT_CL_RCL_ORDER_INCREASING_*.  This allows overlapping
+ * blits to be implemented using the 3D engine.
+ */
+#define VC4_SUBMIT_CL_FIXED_RCL_ORDER			(1 << 1)
+#define VC4_SUBMIT_CL_RCL_ORDER_INCREASING_X		(1 << 2)
+#define VC4_SUBMIT_CL_RCL_ORDER_INCREASING_Y		(1 << 3)
 	__u32 flags;
 
 	/* Returned value of the seqno of this render job (for the
@@ -294,6 +304,7 @@ struct drm_vc4_get_hang_state {
 #define DRM_VC4_PARAM_SUPPORTS_BRANCHES		3
 #define DRM_VC4_PARAM_SUPPORTS_ETC1		4
 #define DRM_VC4_PARAM_SUPPORTS_THREADED_FS	5
+#define DRM_VC4_PARAM_SUPPORTS_FIXED_RCL_ORDER	6
 
 struct drm_vc4_get_param {
 	__u32 param;