From e92be561f1c3fc5219f144605e446966f1e49eb1 Mon Sep 17 00:00:00 2001
From: serge-sans-paille <serge.guelton@telecom-bretagne.eu>
Date: Sat, 14 Dec 2024 22:30:57 +0100
Subject: [PATCH] Fix fma4 support

Fix #1071
---
 include/xsimd/arch/xsimd_fma4.hpp           | 16 ++++++++--------
 include/xsimd/types/xsimd_fma4_register.hpp |  4 ++++
 test/architectures/CMakeLists.txt           |  3 +--
 3 files changed, 13 insertions(+), 10 deletions(-)

diff --git a/include/xsimd/arch/xsimd_fma4.hpp b/include/xsimd/arch/xsimd_fma4.hpp
index e51c7c52a..423331230 100644
--- a/include/xsimd/arch/xsimd_fma4.hpp
+++ b/include/xsimd/arch/xsimd_fma4.hpp
@@ -23,52 +23,52 @@ namespace xsimd
 
         // fnma
         template <class A>
-        XSIMD_INLINE batch<float, A> fnma(simd_register<float, A> const& x, simd_register<float, A> const& y, simd_register<float, A> const& z, requires_arch<fma4>) noexcept
+        XSIMD_INLINE batch<float, A> fnma(batch<float, A> const& x, batch<float, A> const& y, batch<float, A> const& z, requires_arch<fma4>) noexcept
         {
             return _mm_nmacc_ps(x, y, z);
         }
 
         template <class A>
-        XSIMD_INLINE batch<double, A> fnma(simd_register<double, A> const& x, simd_register<double, A> const& y, simd_register<double, A> const& z, requires_arch<fma4>) noexcept
+        XSIMD_INLINE batch<double, A> fnma(batch<double, A> const& x, batch<double, A> const& y, batch<double, A> const& z, requires_arch<fma4>) noexcept
         {
             return _mm_nmacc_pd(x, y, z);
         }
 
         // fnms
         template <class A>
-        XSIMD_INLINE batch<float, A> fnms(simd_register<float, A> const& x, simd_register<float, A> const& y, simd_register<float, A> const& z, requires_arch<fma4>) noexcept
+        XSIMD_INLINE batch<float, A> fnms(batch<float, A> const& x, batch<float, A> const& y, batch<float, A> const& z, requires_arch<fma4>) noexcept
         {
             return _mm_nmsub_ps(x, y, z);
         }
 
         template <class A>
-        XSIMD_INLINE batch<double, A> fnms(simd_register<double, A> const& x, simd_register<double, A> const& y, simd_register<double, A> const& z, requires_arch<fma4>) noexcept
+        XSIMD_INLINE batch<double, A> fnms(batch<double, A> const& x, batch<double, A> const& y, batch<double, A> const& z, requires_arch<fma4>) noexcept
         {
             return _mm_nmsub_pd(x, y, z);
         }
 
         // fma
         template <class A>
-        XSIMD_INLINE batch<float, A> fma(simd_register<float, A> const& x, simd_register<float, A> const& y, simd_register<float, A> const& z, requires_arch<fma4>) noexcept
+        XSIMD_INLINE batch<float, A> fma(batch<float, A> const& x, batch<float, A> const& y, batch<float, A> const& z, requires_arch<fma4>) noexcept
         {
             return _mm_macc_ps(x, y, z);
         }
 
         template <class A>
-        XSIMD_INLINE batch<double, A> fma(simd_register<double, A> const& x, simd_register<double, A> const& y, simd_register<double, A> const& z, requires_arch<fma4>) noexcept
+        XSIMD_INLINE batch<double, A> fma(batch<double, A> const& x, batch<double, A> const& y, batch<double, A> const& z, requires_arch<fma4>) noexcept
         {
             return _mm_macc_pd(x, y, z);
         }
 
         // fms
         template <class A>
-        XSIMD_INLINE batch<float, A> fms(simd_register<float, A> const& x, simd_register<float, A> const& y, simd_register<float, A> const& z, requires_arch<fma4>) noexcept
+        XSIMD_INLINE batch<float, A> fms(batch<float, A> const& x, batch<float, A> const& y, batch<float, A> const& z, requires_arch<fma4>) noexcept
         {
             return _mm_msub_ps(x, y, z);
         }
 
         template <class A>
-        XSIMD_INLINE batch<double, A> fms(simd_register<double, A> const& x, simd_register<double, A> const& y, simd_register<double, A> const& z, requires_arch<fma4>) noexcept
+        XSIMD_INLINE batch<double, A> fms(batch<double, A> const& x, batch<double, A> const& y, batch<double, A> const& z, requires_arch<fma4>) noexcept
         {
             return _mm_msub_pd(x, y, z);
         }
diff --git a/include/xsimd/types/xsimd_fma4_register.hpp b/include/xsimd/types/xsimd_fma4_register.hpp
index 1a066cd20..70f889abb 100644
--- a/include/xsimd/types/xsimd_fma4_register.hpp
+++ b/include/xsimd/types/xsimd_fma4_register.hpp
@@ -14,6 +14,10 @@
 
 #include "./xsimd_sse4_2_register.hpp"
 
+#if XSIMD_WITH_FMA4
+#include <x86intrin.h>
+#endif
+
 namespace xsimd
 {
     /**
diff --git a/test/architectures/CMakeLists.txt b/test/architectures/CMakeLists.txt
index 491f95629..597803904 100644
--- a/test/architectures/CMakeLists.txt
+++ b/test/architectures/CMakeLists.txt
@@ -1,8 +1,7 @@
 set(INTEL_PROCESSORS
-    knl knm skylake-avx512 cannonlake icelake-client
+    bdver1 knl knm skylake-avx512 cannonlake icelake-client
     icelake-server cascadelake cooperlake tigerlake sapphirerapids alderlake
     rocketlake graniterapids graniterapids-d znver4)
-set(CMAKE_CXX_FLAGS "-Werror=unused-command-line-argument")
 
 foreach(INTEL_PROCESSOR ${INTEL_PROCESSORS})
     # Adding the werror here to choke if the -march is incompatible with the