From cfa09d859fb85efcf6765367fa97cad8d6bd3f1c Mon Sep 17 00:00:00 2001
From: Henry de Valence <hdevalence@hdevalence.ca>
Date: Tue, 4 Jun 2019 14:29:33 -0700
Subject: [PATCH] Use upstream IFMA intrinsics now that they exist.

---
 build.rs                         |  9 +--------
 src/backend/vector/ifma/field.rs | 24 +++++++++++++-----------
 src/lib.rs                       |  8 +-------
 3 files changed, 15 insertions(+), 26 deletions(-)

diff --git a/build.rs b/build.rs
index 2d020e1f2..0ffa5fc78 100644
--- a/build.rs
+++ b/build.rs
@@ -1,13 +1,6 @@
-#![cfg_attr(
-    all(feature = "simd_backend", target_feature = "avx512ifma"),
-    feature(simd_ffi)
-)]
-#![cfg_attr(
-    all(feature = "simd_backend", target_feature = "avx512ifma"),
-    feature(link_llvm_intrinsics)
-)]
 #![cfg_attr(all(feature = "alloc", not(feature = "std")), feature(alloc))]
 #![cfg_attr(feature = "nightly", feature(doc_cfg))]
+#![cfg_attr(feature = "simd_backend", feature(stdsimd))]
 #![allow(unused_variables)]
 #![allow(non_snake_case)]
 #![allow(dead_code)]
diff --git a/src/backend/vector/ifma/field.rs b/src/backend/vector/ifma/field.rs
index d220a5a4d..80bd86f1f 100644
--- a/src/backend/vector/ifma/field.rs
+++ b/src/backend/vector/ifma/field.rs
@@ -14,12 +14,18 @@ use packed_simd::{u64x4, IntoBits};
 
 use backend::serial::u64::field::FieldElement51;
 
-#[allow(improper_ctypes)]
-extern "C" {
-    #[link_name = "llvm.x86.avx512.vpmadd52l.uq.256"]
-    fn madd52lo(z: u64x4, x: u64x4, y: u64x4) -> u64x4;
-    #[link_name = "llvm.x86.avx512.vpmadd52h.uq.256"]
-    fn madd52hi(z: u64x4, x: u64x4, y: u64x4) -> u64x4;
+/// A wrapper around `vpmadd52luq` that works on `u64x4`.
+#[inline(always)]
+unsafe fn madd52lo(z: u64x4, x: u64x4, y: u64x4) -> u64x4 {
+    use core::arch::x86_64::_mm256_madd52lo_epu64;
+    _mm256_madd52lo_epu64(z.into_bits(), x.into_bits(), y.into_bits()).into_bits()
+}
+
+/// A wrapper around `vpmadd52huq` that works on `u64x4`.
+#[inline(always)]
+unsafe fn madd52hi(z: u64x4, x: u64x4, y: u64x4) -> u64x4 {
+    use core::arch::x86_64::_mm256_madd52hi_epu64;
+    _mm256_madd52hi_epu64(z.into_bits(), x.into_bits(), y.into_bits()).into_bits()
 }
 
 /// A vector of four field elements in radix 2^51, with unreduced coefficients.
@@ -203,11 +209,7 @@ use subtle::ConditionallySelectable;
 
 impl ConditionallySelectable for F51x4Reduced {
     #[inline]
-    fn conditional_select(
-        a: &F51x4Reduced,
-        b: &F51x4Reduced,
-        choice: Choice,
-    ) -> F51x4Reduced {
+    fn conditional_select(a: &F51x4Reduced, b: &F51x4Reduced, choice: Choice) -> F51x4Reduced {
         let mask = (-(choice.unwrap_u8() as i64)) as u64;
         let mask_vec = u64x4::splat(mask);
         F51x4Reduced([
diff --git a/src/lib.rs b/src/lib.rs
index 07c3befcb..f1ae938d9 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -9,17 +9,11 @@
 // - Henry de Valence <hdevalence@hdevalence.ca>
 
 #![no_std]
-#![cfg_attr(
-    any(
-        all(feature = "simd_backend", target_feature = "avx512ifma"),
-        all(feature = "nightly", rustdoc)
-    ),
-    feature(simd_ffi, link_llvm_intrinsics)
-)]
 #![cfg_attr(feature = "nightly", feature(test))]
 #![cfg_attr(all(feature = "alloc", not(feature = "std")), feature(alloc))]
 #![cfg_attr(feature = "nightly", feature(external_doc))]
 #![cfg_attr(feature = "nightly", feature(doc_cfg))]
+#![cfg_attr(feature = "simd_backend", feature(stdsimd))]
 // Refuse to compile if documentation is missing, but only on nightly.
 //
 // This means that missing docs will still fail CI, but means we can use