freedreno: Move the primitives generated/written updates after the draw.

If we're going to clamp the written value, we need the max vtx update from
the draw's state emit.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9687>

src/gallium/drivers/freedreno/freedreno_draw.c

@@ -352,9 +352,6 @@ fd_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info,
 	batch->back_blit = ctx->in_shadow;
 	batch->num_draws++;
 
-	if (unlikely(ctx->stats_users > 0))
-		update_draw_stats(ctx, info, draws, num_draws);
-
 	/* Clearing last_fence must come after the batch dependency tracking
 	 * (resource_read()/resource_written()), as that can trigger a flush,
 	 * re-populating last_fence
@@ -376,6 +373,9 @@ fd_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info,
 		batch->num_vertices += draws[i].count * info->instance_count;
 	}
 
+	if (unlikely(ctx->stats_users > 0))
+		update_draw_stats(ctx, info, draws, num_draws);
+
 	for (unsigned i = 0; i < ctx->streamout.num_targets; i++) {
 		assert(num_draws == 1);
 		ctx->streamout.offsets[i] += draws[0].count;