Silence Poseidon warnings for ARM targets

plonky2/src/hash/poseidon_goldilocks.rs

@@ -4,8 +4,8 @@
 //! `poseidon_constants.sage` script in the `mir-protocol/hash-constants`
 //! repository.
 
+#[cfg(not(all(target_arch = "aarch64", target_feature = "neon")))]
 use plonky2_field::types::Field;
-use unroll::unroll_for_loops;
 
 use crate::field::goldilocks_field::GoldilocksField;
 use crate::hash::poseidon::{Poseidon, N_PARTIAL_ROUNDS};
@@ -214,9 +214,9 @@ impl Poseidon for GoldilocksField {
          0xdcedab70f40718ba, 0xe796d293a47a64cb, 0x80772dc2645b280b, ],
     ];
 
-    #[cfg(target_arch="x86_64")]
+    #[cfg(not(all(target_arch = "aarch64", target_feature = "neon")))]
     #[inline(always)]
-    #[unroll_for_loops]
+    #[unroll::unroll_for_loops]
     fn mds_layer(state: &[Self; 12]) -> [Self; 12] {
         let mut result = [GoldilocksField::ZERO; 12];
 
@@ -231,8 +231,8 @@ impl Poseidon for GoldilocksField {
             state_l[r] = (s as u32) as u64;
         }
 
-        let state_h = mds_multiply_freq(state_h);
-        let state_l = mds_multiply_freq(state_l);
+        let state_h = poseidon12_mds::mds_multiply_freq(state_h);
+        let state_l = poseidon12_mds::mds_multiply_freq(state_l);
 
         for r in 0..12 {
             let s = state_l[r] as u128 + ((state_h[r] as u128) << 32);
@@ -307,14 +307,15 @@ impl Poseidon for GoldilocksField {
 // MDS layer helper methods
 // The following code has been adapted from winterfell/crypto/src/hash/mds/mds_f64_12x12.rs
 // located at https://github.com/facebook/winterfell.
-
+#[cfg(not(all(target_arch = "aarch64", target_feature = "neon")))]
+mod poseidon12_mds {
     const MDS_FREQ_BLOCK_ONE: [i64; 3] = [16, 32, 16];
     const MDS_FREQ_BLOCK_TWO: [(i64, i64); 3] = [(2, -1), (-4, 1), (16, 1)];
     const MDS_FREQ_BLOCK_THREE: [i64; 3] = [-1, -8, 2];
 
     /// Split 3 x 4 FFT-based MDS vector-multiplication with the Poseidon circulant MDS matrix.
     #[inline(always)]
-fn mds_multiply_freq(state: [u64; 12]) -> [u64; 12] {
+    pub(crate) fn mds_multiply_freq(state: [u64; 12]) -> [u64; 12] {
         let [s0, s1, s2, s3, s4, s5, s6, s7, s8, s9, s10, s11] = state;
 
         let (u0, u1, u2) = fft4_real([s0, s3, s6, s9]);
@@ -341,45 +342,6 @@ fn mds_multiply_freq(state: [u64; 12]) -> [u64; 12] {
         [s0, s1, s2, s3, s4, s5, s6, s7, s8, s9, s10, s11]
     }
 
-/// Real 2-FFT over u64 integers.
-#[inline(always)]
-fn fft2_real(x: [u64; 2]) -> [i64; 2] {
-    [(x[0] as i64 + x[1] as i64), (x[0] as i64 - x[1] as i64)]
-}
-
-/// Real 2-iFFT over u64 integers.
-/// Division by two to complete the inverse FFT is not performed here.
-#[inline(always)]
-fn ifft2_real_unreduced(y: [i64; 2]) -> [u64; 2] {
-    [(y[0] + y[1]) as u64, (y[0] - y[1]) as u64]
-}
-
-/// Real 4-FFT over u64 integers.
-#[inline(always)]
-fn fft4_real(x: [u64; 4]) -> (i64, (i64, i64), i64) {
-    let [z0, z2] = fft2_real([x[0], x[2]]);
-    let [z1, z3] = fft2_real([x[1], x[3]]);
-    let y0 = z0 + z1;
-    let y1 = (z2, -z3);
-    let y2 = z0 - z1;
-    (y0, y1, y2)
-}
-
-/// Real 4-iFFT over u64 integers.
-/// Division by four to complete the inverse FFT is not performed here.
-#[inline(always)]
-fn ifft4_real_unreduced(y: (i64, (i64, i64), i64)) -> [u64; 4] {
-    let z0 = y.0 + y.2;
-    let z1 = y.0 - y.2;
-    let z2 = y.1 .0;
-    let z3 = -y.1 .1;
-
-    let [x0, x2] = ifft2_real_unreduced([z0, z2]);
-    let [x1, x3] = ifft2_real_unreduced([z1, z3]);
-
-    [x0, x1, x2, x3]
-}
-
     #[inline(always)]
     fn block1(x: [i64; 3], y: [i64; 3]) -> [i64; 3] {
         let [x0, x1, x2] = x;
@@ -440,6 +402,46 @@ fn block3(x: [i64; 3], y: [i64; 3]) -> [i64; 3] {
         [z0, z1, z2]
     }
 
+    /// Real 2-FFT over u64 integers.
+    #[inline(always)]
+    pub(crate) fn fft2_real(x: [u64; 2]) -> [i64; 2] {
+        [(x[0] as i64 + x[1] as i64), (x[0] as i64 - x[1] as i64)]
+    }
+
+    /// Real 2-iFFT over u64 integers.
+    /// Division by two to complete the inverse FFT is not performed here.
+    #[inline(always)]
+    pub(crate) fn ifft2_real_unreduced(y: [i64; 2]) -> [u64; 2] {
+        [(y[0] + y[1]) as u64, (y[0] - y[1]) as u64]
+    }
+
+    /// Real 4-FFT over u64 integers.
+    #[inline(always)]
+    pub(crate) fn fft4_real(x: [u64; 4]) -> (i64, (i64, i64), i64) {
+        let [z0, z2] = fft2_real([x[0], x[2]]);
+        let [z1, z3] = fft2_real([x[1], x[3]]);
+        let y0 = z0 + z1;
+        let y1 = (z2, -z3);
+        let y2 = z0 - z1;
+        (y0, y1, y2)
+    }
+
+    /// Real 4-iFFT over u64 integers.
+    /// Division by four to complete the inverse FFT is not performed here.
+    #[inline(always)]
+    pub(crate) fn ifft4_real_unreduced(y: (i64, (i64, i64), i64)) -> [u64; 4] {
+        let z0 = y.0 + y.2;
+        let z1 = y.0 - y.2;
+        let z2 = y.1 .0;
+        let z3 = -y.1 .1;
+
+        let [x0, x2] = ifft2_real_unreduced([z0, z2]);
+        let [x1, x3] = ifft2_real_unreduced([z1, z3]);
+
+        [x0, x1, x2, x3]
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use crate::field::goldilocks_field::GoldilocksField as F;