From 20dbd23b001661e9538d980cbf469732fc63a512 Mon Sep 17 00:00:00 2001
From: Rogiel Sulzbach <rogiel@rogiel.com>
Date: Tue, 26 Dec 2023 15:09:39 -0300
Subject: [PATCH] Fix ARM Neon compilation and MulAdd implementation

---
 include/FastSIMD/ToolSet/ARM/128/f32x4.h | 4 ++--
 include/FastSIMD/ToolSet/ARM/128/m32x4.h | 3 +--
 2 files changed, 3 insertions(+), 4 deletions(-)

diff --git a/include/FastSIMD/ToolSet/ARM/128/f32x4.h b/include/FastSIMD/ToolSet/ARM/128/f32x4.h
index e40f2d4..a0fffbd 100644
--- a/include/FastSIMD/ToolSet/ARM/128/f32x4.h
+++ b/include/FastSIMD/ToolSet/ARM/128/f32x4.h
@@ -232,12 +232,12 @@ namespace FS
     template<FastSIMD::FeatureSet SIMD, typename = EnableIfNative<f32<4, SIMD>>, typename = std::enable_if_t<SIMD & FastSIMD::FeatureFlag::FMA>>
     FS_FORCEINLINE f32<4, SIMD> FMulAdd( const f32<4, SIMD>& a, const f32<4, SIMD>& b, const f32<4, SIMD>& c )
     {
-        return vmlaq_f32( b.native, c.native, a.native );
+        return vmlaq_f32( c.native, b.native, a.native );
     }
 
     template<FastSIMD::FeatureSet SIMD, typename = EnableIfNative<f32<4, SIMD>>, typename = std::enable_if_t<SIMD & FastSIMD::FeatureFlag::FMA>>
     FS_FORCEINLINE f32<4, SIMD> FNMulAdd( const f32<4, SIMD>& a, const f32<4, SIMD>& b, const f32<4, SIMD>& c )
     {
-        return vmlaq_f32( b.native, c.native, a.native );
+        return vmlsq_f32( c.native, b.native, a.native );
     }
 }
diff --git a/include/FastSIMD/ToolSet/ARM/128/m32x4.h b/include/FastSIMD/ToolSet/ARM/128/m32x4.h
index 239a618..5a49230 100644
--- a/include/FastSIMD/ToolSet/ARM/128/m32x4.h
+++ b/include/FastSIMD/ToolSet/ARM/128/m32x4.h
@@ -17,8 +17,7 @@ namespace FS
 
         FS_FORCEINLINE Register() = default;
 
-        template<typename T = NativeType>
-        FS_FORCEINLINE Register( std::enable_if_t<OPTIMISE_FLOAT, T> v ) : native( v ) { }
+        FS_FORCEINLINE Register( NativeType v ) : native( v ) { }
 
         template<typename T = Register<Mask<32, false>, 4, SIMD>>
         FS_FORCEINLINE Register( const std::enable_if_t<OPTIMISE_FLOAT, T>& v ) : native( v.native ) { }