https://bugzilla.mozilla.org/show_bug.cgi?id=1504834#c9
https://bugs.chromium.org/p/skia/issues/detail?id=9781

--- a/gfx/skia/skia/include/private/SkNx.h
+++ b/gfx/skia/skia/include/private/SkNx.h
@@ -238,7 +238,18 @@
     AI SkNx operator*(const SkNx& y) const { return fVal * y.fVal; }
     AI SkNx operator/(const SkNx& y) const { return fVal / y.fVal; }
 
+    // On Big endian the commented out variant doesn't work,
+    // and honestly, I have no idea why it exists in the first place.
+    // The reason its broken is, I think, that it defaults to the double-variant of ToBits()
+    // which gets a 64-bit integer, and FromBits returns 32-bit,
+    // cutting off the wrong half again.
+    // Overall, I see no reason to have ToBits and FromBits at all (even for floats/doubles).
+    // Still we are only "fixing" this for big endian and leave little endian alone (never touch a running system)
+#ifdef SK_CPU_BENDIAN
+    AI SkNx operator&(const SkNx& y) const { return fVal & y.fVal; }
+#else
     AI SkNx operator&(const SkNx& y) const { return FromBits(ToBits(fVal) & ToBits(y.fVal)); }
+#endif
     AI SkNx operator|(const SkNx& y) const { return FromBits(ToBits(fVal) | ToBits(y.fVal)); }
     AI SkNx operator^(const SkNx& y) const { return FromBits(ToBits(fVal) ^ ToBits(y.fVal)); }
 
--- a/gfx/skia/skia/src/opts/SkBlitMask_opts.h
+++ b/gfx/skia/skia/src/opts/SkBlitMask_opts.h
@@ -203,7 +203,13 @@
             //   ~~~>
             // a = 1*aa + d(1-1*aa) = aa + d(1-aa)
             // c = 0*aa + d(1-1*aa) =      d(1-aa)
+
+            // For big endian we have to swap the alpha-mask from 0,0,0,255 to 255,0,0,0
+#ifdef SK_CPU_BENDIAN
+            return Sk4px(Sk16b(aa) & Sk16b(255,0,0,0, 255,0,0,0, 255,0,0,0, 255,0,0,0))
+#else
             return Sk4px(Sk16b(aa) & Sk16b(0,0,0,255, 0,0,0,255, 0,0,0,255, 0,0,0,255))
+#endif
                  + d.approxMulDiv255(aa.inv());
         };
         while (h --> 0) {