Moving mask to const, eliminating transmutes and ensuring const evalu…

…ation and therefore uses the alignment of the destination
mrhooray · Apr 5, 2024 · ac7673c · ac7673c
1 parent 677a3b3
commit ac7673c
Showing 1 changed file with 5 additions and 3 deletions.
diff --git a/src/simd/x86.rs b/src/simd/x86.rs
@@ -49,7 +49,7 @@ impl ValueOps for Value {
         unsafe {
             Self(arch::_mm_xor_si128(
                 arch::_mm_clmulepi64_si128(
-                    arch::_mm_and_si128(self.0, mem::transmute((1u128 << 32) - 1)),
+                    arch::_mm_and_si128(self.0, MASK),
                     x_mod_p.0,
                     0x00,
                 ),
@@ -63,16 +63,18 @@ impl ValueOps for Value {
         // SAFETY: This is only implemented if the target supports sse2, sse4.1, and pclmulqdq
         unsafe {
             let t1 = arch::_mm_clmulepi64_si128(
-                arch::_mm_and_si128(self.0, mem::transmute((1u128 << 32) - 1)),
+                arch::_mm_and_si128(self.0, MASK),
                 px_u.0,
                 0x10,
             );
             let t2 = arch::_mm_clmulepi64_si128(
-                arch::_mm_and_si128(t1, mem::transmute((1u128 << 32) - 1)),
+                arch::_mm_and_si128(t1, MASK),
                 px_u.0,
                 0x00,
             );
             arch::_mm_extract_epi32(arch::_mm_xor_si128(self.0, t2), 1) as u32
         }
     }
 }
+
+const MASK: arch::__m128i = unsafe { mem::transmute((1u128 << 32) - 1) };