Add one AVX512f comparison and the intrinsics needed to test it

Author: Daniel Smith

crates/core_arch/src/simd.rs

@@ -201,3 +201,7 @@ simd_ty!(i32x16[i32]:
 simd_ty!(i64x8[i64]:
          i64, i64, i64, i64, i64, i64, i64, i64
          | x0, x1, x2, x3, x4, x5, x6, x7);
+
+simd_ty!(u64x8[u64]:
+         u64, u64, u64, u64, u64, u64, u64, u64
+         | x0, x1, x2, x3, x4, x5, x6, x7);

crates/core_arch/src/x86/avx512f.rs

@@ -94,6 +94,35 @@ pub unsafe fn _mm512_set1_epi64(a: i64) -> __m512i {
     transmute(i64x8::splat(a))
 }
 
+/// Sets packed 64-bit integers in `dst` with the supplied values.
+///
+/// [Intel's documentation]( https://software.intel.com/sites/landingpage/IntrinsicsGuide/#expand=727,1063,4909,1062,1062,4909&text=_mm512_set_epi64)
+#[inline]
+#[target_feature(enable = "avx512f")]
+pub unsafe fn _mm512_set_epi64(
+    e7: i64,
+    e6: i64,
+    e5: i64,
+    e4: i64,
+    e3: i64,
+    e2: i64,
+    e1: i64,
+    e0: i64,
+) -> __m512i {
+    let r = i64x8(e0, e1, e2, e3, e4, e5, e6, e7);
+    transmute(r)
+}
+
+/// Compare packed unsigned 64-bit integers in a and b for less-than, and store the results in a mask vector.
+///
+/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#expand=727,1063,4909,1062,1062&text=_mm512_cmplt_epu64)
+#[inline]
+#[target_feature(enable = "avx512f")]
+#[cfg_attr(test, assert_instr(vpcmpuq))]
+pub unsafe fn _mm512_cmplt_epu64_mask(a: __m512i, b: __m512i) -> __mmask8 {
+    simd_bitmask::<__m512i, _>(simd_lt(a.as_u64x8(), b.as_u64x8()))
+}
+
 #[cfg(test)]
 mod tests {
     use std;
@@ -197,4 +226,12 @@ mod tests {
         );
         assert_eq_m512i(r, e);
     }
+
+    #[simd_test(enable = "avx512f")]
+    unsafe fn test_mm512_cmplt_epu64_mask() {
+        let a = _mm512_set_epi64(0, 1, -1, u64::MAX as i64, i64::MAX, i64::MIN, 100, -100);
+        let b = _mm512_set1_epi64(-1);
+        let m = _mm512_cmplt_epu64_mask(a, b);
+        assert_eq!(m, 0b11001111);
+    }
 }

crates/core_arch/src/x86/mod.rs

@@ -346,6 +346,10 @@ types! {
 #[allow(non_camel_case_types)]
 pub type __mmask16 = u16;
 
+/// The `__mmask8` type used in AVX-512 intrinsics, a 8-bit integer
+#[allow(non_camel_case_types)]
+pub type __mmask8 = u8;
+
 #[cfg(test)]
 mod test;
 #[cfg(test)]
@@ -509,6 +513,11 @@ pub(crate) trait m512iExt: Sized {
     fn as_i32x16(self) -> crate::core_arch::simd::i32x16 {
         unsafe { transmute(self.as_m512i()) }
     }
+
+    #[inline]
+    fn as_u64x8(self) -> crate::core_arch::simd::u64x8 {
+        unsafe { transmute(self.as_m512i()) }
+    }
 }
 
 impl m512iExt for __m512i {