diff --git a/src/devices/ppu_aarch64.c b/src/devices/ppu_aarch64.c
index b6cf37a..6491336 100644
--- a/src/devices/ppu_aarch64.c
+++ b/src/devices/ppu_aarch64.c
@@ -4,25 +4,21 @@
 void
 ppu_redraw(Ppu *p, Uint32 *screen)
 {
-	/* FIXME(sigrid): do this better */
-	Uint32 *rgba = __builtin_assume_aligned(screen, 16);
-	Uint8 *fg = __builtin_assume_aligned(p->fg.pixels, 16);
-	Uint8 *bg = __builtin_assume_aligned(p->bg.pixels, 16);
-	Uint8 *palette = __builtin_assume_aligned((Uint8*)p->palette, 16);
-	uint8x16x4_t pal = vld4q_u8(palette); enum { R, G, B, A };
+	uint8x16x4_t pal = vld4q_u8((Uint8*)p->palette);
+	Uint8 *fg = p->fg.pixels;
+	Uint8 *bg = p->bg.pixels;
 	int i;
 
-	for(i = 0; i < p->width * p->height; i += 16, fg += 16, bg += 16, rgba += 16) {
+	for(i = 0; i < p->width * p->height; i += 16, fg += 16, bg += 16, screen += 16) {
 		uint8x16_t fg8 = vld1q_u8(fg);
 		uint8x16_t bg8 = vld1q_u8(bg);
-		uint8x16_t bgmask = vceqzq_u8(fg8);
-		uint8x16_t px8 = vorrq_u8(vandq_u8(bg8, bgmask), vandq_u8(fg8, vceqzq_u8(bgmask)));
+		uint8x16_t px8 = vbslq_u8(vceqzq_u8(fg8), bg8, fg8);
 		uint8x16x4_t px = {
-			vqtbl1q_u8(pal.val[R], px8),
-			vqtbl1q_u8(pal.val[G], px8),
-			vqtbl1q_u8(pal.val[B], px8),
-			vqtbl1q_u8(pal.val[A], px8),
+			vqtbl1q_u8(pal.val[0], px8),
+			vqtbl1q_u8(pal.val[1], px8),
+			vqtbl1q_u8(pal.val[2], px8),
+			vdupq_n_u8(0xff),
 		};
-		vst4q_u8((uint8_t*)rgba, px);
+		vst4q_u8((uint8_t*)screen, px);
 	}
 }