src/stdsimd/coresimd/ppsv/codegen/xor.rs

   1 //! Code generation for the xor reduction.
   2 use coresimd::simd::*;
   3
   4 /// LLVM intrinsics used in the xor reduction
   5 #[allow(improper_ctypes)]
   6 extern "C" {
   7     #[link_name = "llvm.experimental.vector.reduce.xor.i8.v2i8"]
   8     fn reduce_xor_i8x2(x: i8x2) -> i8;
   9     #[link_name = "llvm.experimental.vector.reduce.xor.u8.v2u8"]
  10     fn reduce_xor_u8x2(x: u8x2) -> u8;
  11     #[link_name = "llvm.experimental.vector.reduce.xor.i16.v2i16"]
  12     fn reduce_xor_i16x2(x: i16x2) -> i16;
  13     #[link_name = "llvm.experimental.vector.reduce.xor.u16.v2u16"]
  14     fn reduce_xor_u16x2(x: u16x2) -> u16;
  15     #[link_name = "llvm.experimental.vector.reduce.xor.i32.v2i32"]
  16     fn reduce_xor_i32x2(x: i32x2) -> i32;
  17     #[link_name = "llvm.experimental.vector.reduce.xor.u32.v2u32"]
  18     fn reduce_xor_u32x2(x: u32x2) -> u32;
  19     #[link_name = "llvm.experimental.vector.reduce.xor.i64.v2i64"]
  20     fn reduce_xor_i64x2(x: i64x2) -> i64;
  21     #[link_name = "llvm.experimental.vector.reduce.xor.u64.v2u64"]
  22     fn reduce_xor_u64x2(x: u64x2) -> u64;
  23     #[link_name = "llvm.experimental.vector.reduce.xor.i8.v4i8"]
  24     fn reduce_xor_i8x4(x: i8x4) -> i8;
  25     #[link_name = "llvm.experimental.vector.reduce.xor.u8.v4u8"]
  26     fn reduce_xor_u8x4(x: u8x4) -> u8;
  27     #[link_name = "llvm.experimental.vector.reduce.xor.i16.v4i16"]
  28     fn reduce_xor_i16x4(x: i16x4) -> i16;
  29     #[link_name = "llvm.experimental.vector.reduce.xor.u16.v4u16"]
  30     fn reduce_xor_u16x4(x: u16x4) -> u16;
  31     #[link_name = "llvm.experimental.vector.reduce.xor.i32.v4i32"]
  32     fn reduce_xor_i32x4(x: i32x4) -> i32;
  33     #[link_name = "llvm.experimental.vector.reduce.xor.u32.v4u32"]
  34     fn reduce_xor_u32x4(x: u32x4) -> u32;
  35     #[link_name = "llvm.experimental.vector.reduce.xor.i64.v4i64"]
  36     fn reduce_xor_i64x4(x: i64x4) -> i64;
  37     #[link_name = "llvm.experimental.vector.reduce.xor.u64.v4u64"]
  38     fn reduce_xor_u64x4(x: u64x4) -> u64;
  39     #[link_name = "llvm.experimental.vector.reduce.xor.i8.v8i8"]
  40     fn reduce_xor_i8x8(x: i8x8) -> i8;
  41     #[link_name = "llvm.experimental.vector.reduce.xor.u8.v8u8"]
  42     fn reduce_xor_u8x8(x: u8x8) -> u8;
  43     #[link_name = "llvm.experimental.vector.reduce.xor.i16.v8i16"]
  44     fn reduce_xor_i16x8(x: i16x8) -> i16;
  45     #[link_name = "llvm.experimental.vector.reduce.xor.u16.v8u16"]
  46     fn reduce_xor_u16x8(x: u16x8) -> u16;
  47     #[link_name = "llvm.experimental.vector.reduce.xor.i32.v8i32"]
  48     fn reduce_xor_i32x8(x: i32x8) -> i32;
  49     #[link_name = "llvm.experimental.vector.reduce.xor.u32.v8u32"]
  50     fn reduce_xor_u32x8(x: u32x8) -> u32;
  51     #[link_name = "llvm.experimental.vector.reduce.xor.i64.v8i64"]
  52     fn reduce_xor_i64x8(x: i64x8) -> i64;
  53     #[link_name = "llvm.experimental.vector.reduce.xor.u64.v8u64"]
  54     fn reduce_xor_u64x8(x: u64x8) -> u64;
  55     #[link_name = "llvm.experimental.vector.reduce.xor.i8.v16i8"]
  56     fn reduce_xor_i8x16(x: i8x16) -> i8;
  57     #[link_name = "llvm.experimental.vector.reduce.xor.u8.v16u8"]
  58     fn reduce_xor_u8x16(x: u8x16) -> u8;
  59     #[link_name = "llvm.experimental.vector.reduce.xor.i16.v16i16"]
  60     fn reduce_xor_i16x16(x: i16x16) -> i16;
  61     #[link_name = "llvm.experimental.vector.reduce.xor.u16.v16u16"]
  62     fn reduce_xor_u16x16(x: u16x16) -> u16;
  63     #[link_name = "llvm.experimental.vector.reduce.xor.i32.v16i32"]
  64     fn reduce_xor_i32x16(x: i32x16) -> i32;
  65     #[link_name = "llvm.experimental.vector.reduce.xor.u32.v16u32"]
  66     fn reduce_xor_u32x16(x: u32x16) -> u32;
  67     #[link_name = "llvm.experimental.vector.reduce.xor.i8.v32i8"]
  68     fn reduce_xor_i8x32(x: i8x32) -> i8;
  69     #[link_name = "llvm.experimental.vector.reduce.xor.u8.v32u8"]
  70     fn reduce_xor_u8x32(x: u8x32) -> u8;
  71     #[link_name = "llvm.experimental.vector.reduce.xor.i16.v32i16"]
  72     fn reduce_xor_i16x32(x: i16x32) -> i16;
  73     #[link_name = "llvm.experimental.vector.reduce.xor.u16.v32u16"]
  74     fn reduce_xor_u16x32(x: u16x32) -> u16;
  75     #[link_name = "llvm.experimental.vector.reduce.xor.i8.v64i8"]
  76     fn reduce_xor_i8x64(x: i8x64) -> i8;
  77     #[link_name = "llvm.experimental.vector.reduce.xor.u8.v64u8"]
  78     fn reduce_xor_u8x64(x: u8x64) -> u8;
  79 }
  80
  81 /// Reduction: horizontal bitwise xor of the vector elements.
  82 #[cfg_attr(feature = "cargo-clippy", allow(stutter))]
  83 pub trait ReduceXor {
  84     /// Result type of the reduction.
  85     type Acc;
  86     /// Computes the horizontal bitwise xor of the vector elements.
  87     fn reduce_xor(self) -> Self::Acc;
  88 }
  89
  90 macro_rules! red_xor {
  91     ($id:ident, $elem_ty:ident, $llvm_intr:ident) => {
  92         impl ReduceXor for $id {
  93             type Acc = $elem_ty;
  94             #[cfg(not(target_arch = "aarch64"))]
  95             #[inline]
  96             fn reduce_xor(self) -> Self::Acc {
  97                 unsafe { $llvm_intr(self.into_bits()) }
  98             }
  99             // FIXME: broken in AArch64
 100             #[cfg(target_arch = "aarch64")]
 101             #[inline]
 102             fn reduce_xor(self) -> Self::Acc {
 103                 let mut x = self.extract(0) as Self::Acc;
 104                 for i in 1..$id::lanes() {
 105                     x ^= self.extract(i) as Self::Acc;
 106                 }
 107                 x
 108             }
 109         }
 110     };
 111 }
 112 red_xor!(i8x2, i8, reduce_xor_i8x2);
 113 red_xor!(u8x2, u8, reduce_xor_u8x2);
 114 red_xor!(i16x2, i16, reduce_xor_i16x2);
 115 red_xor!(u16x2, u16, reduce_xor_u16x2);
 116 red_xor!(i32x2, i32, reduce_xor_i32x2);
 117 red_xor!(u32x2, u32, reduce_xor_u32x2);
 118 red_xor!(i64x2, i64, reduce_xor_i64x2);
 119 red_xor!(u64x2, u64, reduce_xor_u64x2);
 120 red_xor!(i8x4, i8, reduce_xor_i8x4);
 121 red_xor!(u8x4, u8, reduce_xor_u8x4);
 122 red_xor!(i16x4, i16, reduce_xor_i16x4);
 123 red_xor!(u16x4, u16, reduce_xor_u16x4);
 124 red_xor!(i32x4, i32, reduce_xor_i32x4);
 125 red_xor!(u32x4, u32, reduce_xor_u32x4);
 126 red_xor!(i64x4, i64, reduce_xor_i64x4);
 127 red_xor!(u64x4, u64, reduce_xor_u64x4);
 128 red_xor!(i8x8, i8, reduce_xor_i8x8);
 129 red_xor!(u8x8, u8, reduce_xor_u8x8);
 130 red_xor!(i16x8, i16, reduce_xor_i16x8);
 131 red_xor!(u16x8, u16, reduce_xor_u16x8);
 132 red_xor!(i32x8, i32, reduce_xor_i32x8);
 133 red_xor!(u32x8, u32, reduce_xor_u32x8);
 134 red_xor!(i64x8, i64, reduce_xor_i64x8);
 135 red_xor!(u64x8, u64, reduce_xor_u64x8);
 136 red_xor!(i8x16, i8, reduce_xor_i8x16);
 137 red_xor!(u8x16, u8, reduce_xor_u8x16);
 138 red_xor!(i16x16, i16, reduce_xor_i16x16);
 139 red_xor!(u16x16, u16, reduce_xor_u16x16);
 140 red_xor!(i32x16, i32, reduce_xor_i32x16);
 141 red_xor!(u32x16, u32, reduce_xor_u32x16);
 142 red_xor!(i8x32, i8, reduce_xor_i8x32);
 143 red_xor!(u8x32, u8, reduce_xor_u8x32);
 144 red_xor!(i16x32, i16, reduce_xor_i16x32);
 145 red_xor!(u16x32, u16, reduce_xor_u16x32);
 146 red_xor!(i8x64, i8, reduce_xor_i8x64);
 147 red_xor!(u8x64, u8, reduce_xor_u8x64);
 148
 149 red_xor!(b8x2, i8, reduce_xor_i8x2);
 150 red_xor!(b8x4, i8, reduce_xor_i8x4);
 151 red_xor!(b8x8, i8, reduce_xor_i8x8);
 152 red_xor!(b8x16, i8, reduce_xor_i8x16);
 153 red_xor!(b8x32, i8, reduce_xor_i8x32);
 154 red_xor!(b8x64, i8, reduce_xor_i8x64);
 155
 156 #[cfg(test)]
 157 mod tests {
 158     use super::ReduceXor;
 159     use coresimd::simd::*;
 160
 161     // note: these are tested in the portable vector API tests
 162
 163     #[test]
 164     fn reduce_xor_i32x4() {
 165         let v = i32x4::splat(1);
 166         assert_eq!(v.reduce_xor(), 0_i32);
 167         let v = i32x4::new(1, 0, 0, 0);
 168         assert_eq!(v.reduce_xor(), 1_i32);
 169     }
 170 }