Add _mm512_mask_i32gather_epi64

Author: Daniel Smith

crates/core_arch/src/x86/avx512f.rs

@@ -107,6 +107,32 @@ pub unsafe fn _mm512_i32gather_epi64(offsets: __m256i, slice: *const u8, scale:
     transmute(r)
 }
 
+/// Gather 64-bit integers from memory using 32-bit indices.
+///
+/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_mask_i32gather_epi64)
+#[inline]
+#[target_feature(enable = "avx512f")]
+#[cfg_attr(test, assert_instr(vpgatherdq, scale = 1))]
+pub unsafe fn _mm512_mask_i32gather_epi64(
+    src: __m512i,
+    mask: __mmask8,
+    offsets: __m256i,
+    slice: *const u8,
+    scale: i32,
+) -> __m512i {
+    let src = src.as_i64x8();
+    let mask = mask as i8;
+    let slice = slice as *const i8;
+    let offsets = offsets.as_i32x8();
+    macro_rules! call {
+        ($imm8:expr) => {
+            vpgatherdq(src, slice, offsets, mask, $imm8)
+        };
+    }
+    let r = constify_imm8!(scale, call);
+    transmute(r)
+}
+
 #[allow(improper_ctypes)]
 extern "C" {
     #[link_name = "llvm.x86.avx512.gather.dpq.512"]

crates/core_arch/src/x86_64/avx512f.rs

@@ -235,7 +235,6 @@ mod tests {
         assert_eq_m512i(r, _mm512_setr_epi64(7, 6, 5, 4, 3, 2, 1, 0))
     }
 
-    // _mm512_i32gather_epi64(offsets: __m256i, slice: *const i64, scale: i32)
     #[simd_test(enable = "avx512f")]
     unsafe fn test_mm512_i32gather_epi64() {
         let mut arr = [0i64; 128];
@@ -248,4 +247,19 @@ mod tests {
         let r = _mm512_i32gather_epi64(index, arr.as_ptr() as *const u8, 8);
         assert_eq_m512i(r, _mm512_setr_epi64(0, 16, 32, 48, 64, 80, 96, 112));
     }
+
+    #[simd_test(enable = "avx512f")]
+    unsafe fn test_mm512_mask_i32gather_epi64() {
+        let mut arr = [0i64; 128];
+        for i in 0..128i64 {
+            arr[i as usize] = i;
+        }
+        let src = _mm512_set1_epi64(2);
+        let mask = 0b10101010;
+        #[rustfmt::skip]
+        let index = _mm256_setr_epi32(0, 16, 32, 48, 64, 80, 96, 112);
+        // A multiplier of 8 is word-addressing
+        let r = _mm512_mask_i32gather_epi64(src, mask, index, arr.as_ptr() as *const u8, 8);
+        assert_eq_m512i(r, _mm512_setr_epi64(2, 16, 2, 48, 2, 80, 2, 112));
+    }
 }