Use packed_simd::shuffle instead of vqtbx1q_u8

dalek-cryptography · Dec 8, 2022 · 343be3a · 343be3a
1 parent d60beaa
commit 343be3a
Showing 1 changed file with 6 additions and 11 deletions.
diff --git a/src/backend/vector/neon/field.rs b/src/backend/vector/neon/field.rs
@@ -212,33 +212,28 @@ impl FieldElement2625x4 {
         #[inline(always)]
         fn blend_lanes(x: (u32x4, u32x4), y: (u32x4, u32x4), control: Lanes) -> (u32x4, u32x4) {
             unsafe {
-                use core::arch::aarch64::vqtbx1q_u8;
+                use packed_simd::shuffle;
                 match control {
                     Lanes::C => {
-                        (x.0,
-                         vqtbx1q_u8(x.1.into_bits(), y.1.into_bits(), u8x16::new( 0,  1,  2,  3, 16, 16, 16, 16,  8,  9, 10, 11, 16, 16, 16, 16).into_bits()).into_bits())
+                        (x.0, shuffle!(y.1, x.1, [0, 5, 2, 7]))
                     }
                     Lanes::D => {
-                        (x.0,
-                         vqtbx1q_u8(x.1.into_bits(), y.1.into_bits(), u8x16::new(16, 16, 16, 16,  4,  5,  6,  7, 16, 16, 16, 16, 12, 13, 14, 15).into_bits()).into_bits())
+                        (x.0, shuffle!(y.1, x.1, [4, 1, 6, 3]))
                     }
                     Lanes::AD => {
-                        (vqtbx1q_u8(x.0.into_bits(), y.0.into_bits(), u8x16::new( 0,  1,  2,  3, 16, 16, 16, 16,  8,  9, 10, 11, 16, 16, 16, 16).into_bits() ).into_bits(),
-                         vqtbx1q_u8(x.1.into_bits(), y.1.into_bits(), u8x16::new(16, 16, 16, 16,  4,  5,  6,  7, 16, 16, 16, 16, 12, 13, 14, 15).into_bits() ).into_bits())
+                        (shuffle!(y.0, x.0, [0, 5, 2, 7]), shuffle!(y.1, x.1, [4, 1, 6, 3]))
                     }
                     Lanes::AB => {
                         (y.0, x.1)
                     }
                     Lanes::AC => {
-                        (vqtbx1q_u8(x.0.into_bits(), y.0.into_bits(), u8x16::new( 0,  1,  2,  3, 16, 16, 16, 16,  8,  9, 10, 11, 16, 16, 16, 16).into_bits()).into_bits(),
-                         vqtbx1q_u8(x.1.into_bits(), y.1.into_bits(), u8x16::new( 0,  1,  2,  3, 16, 16, 16, 16,  8,  9, 10, 11, 16, 16, 16, 16).into_bits()).into_bits())
+                        (shuffle!(y.0, x.0, [0, 5, 2, 7]), shuffle!(y.1, x.1, [0, 5, 2, 7]))
                     }
                     Lanes::CD => {
                         (x.0, y.1)    
                     }
                     Lanes::BC => {
-                        (vqtbx1q_u8(x.0.into_bits(), y.0.into_bits(), u8x16::new(16, 16, 16, 16,  4,  5,  6,  7, 16, 16, 16, 16, 12, 13, 14, 15).into_bits() ).into_bits(),
-                         vqtbx1q_u8(x.1.into_bits(), y.1.into_bits(), u8x16::new( 0,  1,  2,  3, 16, 16, 16, 16,  8,  9, 10, 11, 16, 16, 16, 16).into_bits() ).into_bits())
+                        (shuffle!(y.0, x.0, [4, 1, 6, 3]), shuffle!(y.1, x.1, [0, 5, 2, 7]))
                     }
                     Lanes::ABCD => {
                         y