From c2f59483f96cf1ab1e92cf10e0f9094432a8374c Mon Sep 17 00:00:00 2001
From: Jubilee Young <workingjubilee@gmail.com>
Date: Tue, 21 Sep 2021 18:55:05 -0700
Subject: [PATCH] Feature-flag fused mul-add to block libcalls

---
 crates/core_simd/src/intrinsics.rs   | 13 +++++------
 crates/core_simd/src/vector/float.rs |  1 +
 crates/core_simd/tests/ops_macros.rs | 33 +++++++++++++++-------------
 3 files changed, 25 insertions(+), 22 deletions(-)

diff --git a/crates/core_simd/src/intrinsics.rs b/crates/core_simd/src/intrinsics.rs
index 3ed9845d608e..5783950f353e 100644
--- a/crates/core_simd/src/intrinsics.rs
+++ b/crates/core_simd/src/intrinsics.rs
@@ -46,13 +46,6 @@ extern "platform-intrinsic" {
     /// fabs
     pub(crate) fn simd_fabs<T>(x: T) -> T;
 
-    /// fsqrt
-    #[cfg(feature = "std")]
-    pub(crate) fn simd_fsqrt<T>(x: T) -> T;
-
-    /// fma
-    pub(crate) fn simd_fma<T>(x: T, y: T, z: T) -> T;
-
     pub(crate) fn simd_eq<T, U>(x: T, y: T) -> U;
     pub(crate) fn simd_ne<T, U>(x: T, y: T) -> U;
     pub(crate) fn simd_lt<T, U>(x: T, y: T) -> U;
@@ -110,6 +103,12 @@ mod std {
 
         // trunc
         pub(crate) fn simd_trunc<T>(x: T) -> T;
+
+        // fsqrt
+        pub(crate) fn simd_fsqrt<T>(x: T) -> T;
+
+        // fma
+        pub(crate) fn simd_fma<T>(x: T, y: T, z: T) -> T;
     }
 }
 
diff --git a/crates/core_simd/src/vector/float.rs b/crates/core_simd/src/vector/float.rs
index 21a6c43e153a..c09d0ac84d24 100644
--- a/crates/core_simd/src/vector/float.rs
+++ b/crates/core_simd/src/vector/float.rs
@@ -42,6 +42,7 @@ macro_rules! impl_float_vector {
             /// architecture has a dedicated `fma` CPU instruction.  However, this is not always
             /// true, and will be heavily dependent on designing algorithms with specific target
             /// hardware in mind.
+            #[cfg(feature = "std")]
             #[inline]
             pub fn mul_add(self, a: Self, b: Self) -> Self {
                 unsafe { intrinsics::simd_fma(self, a, b) }
diff --git a/crates/core_simd/tests/ops_macros.rs b/crates/core_simd/tests/ops_macros.rs
index 0c45ea2367c9..31b7ee206959 100644
--- a/crates/core_simd/tests/ops_macros.rs
+++ b/crates/core_simd/tests/ops_macros.rs
@@ -437,14 +437,6 @@ macro_rules! impl_float_tests {
                     )
                 }
 
-                fn mul_add<const LANES: usize>() {
-                    test_helpers::test_ternary_elementwise(
-                        &Vector::<LANES>::mul_add,
-                        &Scalar::mul_add,
-                        &|_, _, _| true,
-                    )
-                }
-
                 fn recip<const LANES: usize>() {
                     test_helpers::test_unary_elementwise(
                         &Vector::<LANES>::recip,
@@ -601,13 +593,24 @@ macro_rules! impl_float_tests {
             }
 
             #[cfg(feature = "std")]
-            test_helpers::test_lanes! {
-                fn sqrt<const LANES: usize>() {
-                    test_helpers::test_unary_elementwise(
-                        &Vector::<LANES>::sqrt,
-                        &Scalar::sqrt,
-                        &|_| true,
-                    )
+            mod std {
+                use super::*;
+                test_helpers::test_lanes! {
+                    fn sqrt<const LANES: usize>() {
+                        test_helpers::test_unary_elementwise(
+                            &Vector::<LANES>::sqrt,
+                            &Scalar::sqrt,
+                            &|_| true,
+                        )
+                    }
+
+                    fn mul_add<const LANES: usize>() {
+                        test_helpers::test_ternary_elementwise(
+                            &Vector::<LANES>::mul_add,
+                            &Scalar::mul_add,
+                            &|_, _, _| true,
+                        )
+                    }
                 }
             }
         }