Add 64 bit integer AVX512f comparisons and the intrinsics needed to test them

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 +94,132 @@ pub unsafe fn _mm512_set1_epi64(a: i64) -> __m512i {
     transmute(i64x8::splat(a))
 }
 
+/// 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(vpcmp))]
+pub unsafe fn _mm512_cmplt_epu64_mask(a: __m512i, b: __m512i) -> __mmask8 {
+    simd_bitmask::<__m512i, _>(simd_lt(a.as_u64x8(), b.as_u64x8()))
+}
+
+///Compare packed unsigned 64-bit integers in a and b for less-than, and store the results in a mask vector k
+/// using zeromask m (elements are zeroed out when the corresponding mask bit is not set).
+///
+/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#expand=727,1063,4909,1062,1062,1063&text=_mm512_mask_cmplt_epu64)
+#[inline]
+#[target_feature(enable = "avx512f")]
+#[cfg_attr(test, assert_instr(vpcmp))]
+pub unsafe fn _mm512_mask_cmplt_epu64_mask(m: __mmask8, a: __m512i, b: __m512i) -> __mmask8 {
+    _mm512_cmplt_epu64_mask(a, b) & m
+}
+
+/// 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_cmpgt_epu64)
+#[inline]
+#[target_feature(enable = "avx512f")]
+#[cfg_attr(test, assert_instr(vpcmp))]
+pub unsafe fn _mm512_cmpgt_epu64_mask(a: __m512i, b: __m512i) -> __mmask8 {
+    simd_bitmask::<__m512i, _>(simd_gt(a.as_u64x8(), b.as_u64x8()))
+}
+
+///Compare packed unsigned 64-bit integers in a and b for less-than, and store the results in a mask vector k
+/// using zeromask m (elements are zeroed out when the corresponding mask bit is not set).
+///
+/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#expand=727,1063,4909,1062,1062,1063&text=_mm512_mask_cmpgt_epu64)
+#[inline]
+#[target_feature(enable = "avx512f")]
+#[cfg_attr(test, assert_instr(vpcmp))]
+pub unsafe fn _mm512_mask_cmpgt_epu64_mask(m: __mmask8, a: __m512i, b: __m512i) -> __mmask8 {
+    _mm512_cmpgt_epu64_mask(a, b) & m
+}
+
+/// 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_cmpeq_epu64)
+#[inline]
+#[target_feature(enable = "avx512f")]
+#[cfg_attr(test, assert_instr(vpcmp))]
+pub unsafe fn _mm512_cmpeq_epu64_mask(a: __m512i, b: __m512i) -> __mmask8 {
+    simd_bitmask::<__m512i, _>(simd_eq(a.as_u64x8(), b.as_u64x8()))
+}
+
+///Compare packed unsigned 64-bit integers in a and b for less-than, and store the results in a mask vector k
+/// using zeromask m (elements are zeroed out when the corresponding mask bit is not set).
+///
+/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#expand=727,1063,4909,1062,1062,1063&text=_mm512_mask_cmpeq_epu64)
+#[inline]
+#[target_feature(enable = "avx512f")]
+#[cfg_attr(test, assert_instr(vpcmp))]
+pub unsafe fn _mm512_mask_cmpeq_epu64_mask(m: __mmask8, a: __m512i, b: __m512i) -> __mmask8 {
+    _mm512_cmpeq_epu64_mask(a, b) & m
+}
+
+/// 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_epi64)
+#[inline]
+#[target_feature(enable = "avx512f")]
+#[cfg_attr(test, assert_instr(vpcmp))]
+pub unsafe fn _mm512_cmplt_epi64_mask(a: __m512i, b: __m512i) -> __mmask8 {
+    simd_bitmask::<__m512i, _>(simd_lt(a.as_i64x8(), b.as_i64x8()))
+}
+
+///Compare packed unsigned 64-bit integers in a and b for less-than, and store the results in a mask vector k
+/// using zeromask m (elements are zeroed out when the corresponding mask bit is not set).
+///
+/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#expand=727,1063,4909,1062,1062,1063&text=_mm512_mask_cmplt_epi64)
+#[inline]
+#[target_feature(enable = "avx512f")]
+#[cfg_attr(test, assert_instr(vpcmp))]
+pub unsafe fn _mm512_mask_cmplt_epi64_mask(m: __mmask8, a: __m512i, b: __m512i) -> __mmask8 {
+    _mm512_cmplt_epi64_mask(a, b) & m
+}
+
+/// 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_cmpgt_epi64)
+#[inline]
+#[target_feature(enable = "avx512f")]
+#[cfg_attr(test, assert_instr(vpcmp))]
+pub unsafe fn _mm512_cmpgt_epi64_mask(a: __m512i, b: __m512i) -> __mmask8 {
+    simd_bitmask::<__m512i, _>(simd_gt(a.as_i64x8(), b.as_i64x8()))
+}
+
+///Compare packed unsigned 64-bit integers in a and b for less-than, and store the results in a mask vector k
+/// using zeromask m (elements are zeroed out when the corresponding mask bit is not set).
+///
+/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#expand=727,1063,4909,1062,1062,1063&text=_mm512_mask_cmpgt_epi64)
+#[inline]
+#[target_feature(enable = "avx512f")]
+#[cfg_attr(test, assert_instr(vpcmp))]
+pub unsafe fn _mm512_mask_cmpgt_epi64_mask(m: __mmask8, a: __m512i, b: __m512i) -> __mmask8 {
+    _mm512_cmpgt_epi64_mask(a, b) & m
+}
+
+/// 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_cmpeq_epi64)
+#[inline]
+#[target_feature(enable = "avx512f")]
+#[cfg_attr(test, assert_instr(vpcmp))]
+pub unsafe fn _mm512_cmpeq_epi64_mask(a: __m512i, b: __m512i) -> __mmask8 {
+    simd_bitmask::<__m512i, _>(simd_eq(a.as_i64x8(), b.as_i64x8()))
+}
+
+///Compare packed unsigned 64-bit integers in a and b for less-than, and store the results in a mask vector k
+/// using zeromask m (elements are zeroed out when the corresponding mask bit is not set).
+///
+/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#expand=727,1063,4909,1062,1062,1063&text=_mm512_mask_cmpeq_epi64)
+#[inline]
+#[target_feature(enable = "avx512f")]
+#[cfg_attr(test, assert_instr(vpcmp))]
+pub unsafe fn _mm512_mask_cmpeq_epi64_mask(m: __mmask8, a: __m512i, b: __m512i) -> __mmask8 {
+    _mm512_cmpeq_epi64_mask(a, b) & m
+}
+
 #[cfg(test)]
 mod tests {
     use std;

crates/core_arch/src/x86/mod.rs

@@ -346,6 +346,10 @@ types! {
 #[allow(non_camel_case_types)]
 pub type __mmask16 = i16;
 
+/// 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 +513,16 @@ 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()) }
+    }
+
+    #[inline]
+    fn as_i64x8(self) -> crate::core_arch::simd::i64x8 {
+        unsafe { transmute(self.as_m512i()) }
+    }
 }
 
 impl m512iExt for __m512i {

crates/core_arch/src/x86_64/avx512f.rs

@@ -0,0 +1,165 @@
+use crate::{
+    core_arch::{simd::*, x86::*},
+    mem::transmute,
+};
+
+/// 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(
+    e0: i64,
+    e1: i64,
+    e2: i64,
+    e3: i64,
+    e4: i64,
+    e5: i64,
+    e6: i64,
+    e7: i64,
+) -> __m512i {
+    _mm512_setr_epi64(e7, e6, e5, e4, e3, e2, e1, e0)
+}
+
+/// Sets packed 64-bit integers in `dst` with the supplied values in
+/// reverse order.
+///
+/// [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_setr_epi64(
+    e0: i64,
+    e1: i64,
+    e2: i64,
+    e3: i64,
+    e4: i64,
+    e5: i64,
+    e6: i64,
+    e7: i64,
+) -> __m512i {
+    let r = i64x8::new(e0, e1, e2, e3, e4, e5, e6, e7);
+    transmute(r)
+}
+
+#[cfg(test)]
+mod tests {
+    use std;
+    use stdarch_test::simd_test;
+
+    use crate::core_arch::x86::*;
+    use crate::core_arch::x86_64::*;
+
+    #[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);
+    }
+
+    #[simd_test(enable = "avx512f")]
+    unsafe fn test_mm512_mask_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 mask = 0b01111010;
+        let r = _mm512_mask_cmplt_epu64_mask(mask, a, b);
+        assert_eq!(r, 0b01001010);
+    }
+
+    #[simd_test(enable = "avx512f")]
+    unsafe fn test_mm512_cmpgt_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_cmpgt_epu64_mask(b, a);
+        assert_eq!(m, 0b11001111);
+    }
+
+    #[simd_test(enable = "avx512f")]
+    unsafe fn test_mm512_mask_cmpgt_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 mask = 0b01111010;
+        let r = _mm512_mask_cmpgt_epu64_mask(mask, b, a);
+        assert_eq!(r, 0b01001010);
+    }
+
+    #[simd_test(enable = "avx512f")]
+    unsafe fn test_mm512_cmpeq_epu64_mask() {
+        let a = _mm512_set_epi64(0, 1, -1, u64::MAX as i64, i64::MAX, i64::MIN, 100, -100);
+        let b = _mm512_set_epi64(0, 1, 13, 42, i64::MAX, i64::MIN, 100, -100);
+        let m = _mm512_cmpeq_epu64_mask(b, a);
+        assert_eq!(m, 0b11001111);
+    }
+
+    #[simd_test(enable = "avx512f")]
+    unsafe fn test_mm512_mask_cmpeq_epu64_mask() {
+        let a = _mm512_set_epi64(0, 1, -1, u64::MAX as i64, i64::MAX, i64::MIN, 100, -100);
+        let b = _mm512_set_epi64(0, 1, 13, 42, i64::MAX, i64::MIN, 100, -100);
+        let mask = 0b01111010;
+        let r = _mm512_mask_cmpeq_epu64_mask(mask, b, a);
+        assert_eq!(r, 0b01001010);
+    }
+
+    #[simd_test(enable = "avx512f")]
+    unsafe fn test_mm512_cmplt_epi64_mask() {
+        let a = _mm512_set_epi64(0, 1, -1, 13, i64::MAX, i64::MIN, 100, -100);
+        let b = _mm512_set1_epi64(-1);
+        let m = _mm512_cmplt_epi64_mask(a, b);
+        assert_eq!(m, 0b00000101);
+    }
+
+    #[simd_test(enable = "avx512f")]
+    unsafe fn test_mm512_mask_cmplt_epi64_mask() {
+        let a = _mm512_set_epi64(0, 1, -1, 13, i64::MAX, i64::MIN, 100, -100);
+        let b = _mm512_set1_epi64(-1);
+        let mask = 0b01100110;
+        let r = _mm512_mask_cmplt_epi64_mask(mask, a, b);
+        assert_eq!(r, 0b00000100);
+    }
+
+    #[simd_test(enable = "avx512f")]
+    unsafe fn test_mm512_cmpgt_epi64_mask() {
+        let a = _mm512_set_epi64(0, 1, -1, 13, i64::MAX, i64::MIN, 100, -100);
+        let b = _mm512_set1_epi64(-1);
+        let m = _mm512_cmpgt_epi64_mask(b, a);
+        assert_eq!(m, 0b00000101);
+    }
+
+    #[simd_test(enable = "avx512f")]
+    unsafe fn test_mm512_mask_cmpgt_epi64_mask() {
+        let a = _mm512_set_epi64(0, 1, -1, 13, i64::MAX, i64::MIN, 100, -100);
+        let b = _mm512_set1_epi64(-1);
+        let mask = 0b01100110;
+        let r = _mm512_mask_cmpgt_epi64_mask(mask, b, a);
+        assert_eq!(r, 0b00000100);
+    }
+
+    #[simd_test(enable = "avx512f")]
+    unsafe fn test_mm512_cmpeq_epi64_mask() {
+        let a = _mm512_set_epi64(0, 1, -1, 13, i64::MAX, i64::MIN, 100, -100);
+        let b = _mm512_set_epi64(0, 1, 13, 42, i64::MAX, i64::MIN, 100, -100);
+        let m = _mm512_cmpeq_epi64_mask(b, a);
+        assert_eq!(m, 0b11001111);
+    }
+
+    #[simd_test(enable = "avx512f")]
+    unsafe fn test_mm512_mask_cmpeq_epi64_mask() {
+        let a = _mm512_set_epi64(0, 1, -1, 13, i64::MAX, i64::MIN, 100, -100);
+        let b = _mm512_set_epi64(0, 1, 13, 42, i64::MAX, i64::MIN, 100, -100);
+        let mask = 0b01111010;
+        let r = _mm512_mask_cmpeq_epi64_mask(mask, b, a);
+        assert_eq!(r, 0b01001010);
+    }
+
+    #[simd_test(enable = "avx512f")]
+    unsafe fn test_mm512_set_epi64() {
+        let r = _mm512_setr_epi64(0, 1, 2, 3, 4, 5, 6, 7);
+        assert_eq_m512i(r, _mm512_set_epi64(7, 6, 5, 4, 3, 2, 1, 0))
+    }
+
+    #[simd_test(enable = "avx512f")]
+    unsafe fn test_mm512_setr_epi64() {
+        let r = _mm512_set_epi64(0, 1, 2, 3, 4, 5, 6, 7);
+        assert_eq_m512i(r, _mm512_setr_epi64(7, 6, 5, 4, 3, 2, 1, 0))
+    }
+}

crates/core_arch/src/x86_64/mod.rs

@@ -33,6 +33,9 @@ pub use self::bmi2::*;
 mod avx2;
 pub use self::avx2::*;
 
+mod avx512f;
+pub use self::avx512f::*;
+
 mod bswap;
 pub use self::bswap::*;
 

crates/stdarch-verify/src/lib.rs

@@ -145,6 +145,7 @@ fn to_type(t: &syn::Type) -> proc_macro2::TokenStream {
             "__m512" => quote! { &M512 },
             "__m512d" => quote! { &M512D },
             "__m512i" => quote! { &M512I },
+            "__mmask8" => quote! { &MMASK8 },
             "__mmask16" => quote! { &MMASK16 },
             "__m64" => quote! { &M64 },
             "bool" => quote! { &BOOL },

crates/stdarch-verify/tests/x86-intel.rs

@@ -53,6 +53,7 @@ static M256D: Type = Type::M256D;
 static M512: Type = Type::M512;
 static M512I: Type = Type::M512I;
 static M512D: Type = Type::M512D;
+static MMASK8: Type = Type::MMASK8;
 static MMASK16: Type = Type::MMASK16;
 
 static TUPLE: Type = Type::Tuple;
@@ -76,6 +77,7 @@ enum Type {
     M512,
     M512D,
     M512I,
+    MMASK8,
     MMASK16,
     Tuple,
     CpuidResult,
@@ -653,6 +655,7 @@ fn equate(t: &Type, intel: &str, intrinsic: &str, is_const: bool) -> Result<(),
         (&Type::ConstPtr(&Type::M512I), "__m512i const*") => {}
         (&Type::ConstPtr(&Type::M512D), "__m512d const*") => {}
 
+        (&Type::MMASK8, "__mmask8") => {}
         (&Type::MMASK16, "__mmask16") => {}
 
         // This is a macro (?) in C which seems to mutate its arguments, but