add avx2 support for phi coprocessor

CMakeModules/checkAVX2pe.cpp

+
+#include "immintrin.h"
+
+
+int main() {
+#ifdef __MIC__
+	__m512 tx, ty ;
+	tx += ty ;
+#endif
+  return 0;
+}
+

CMakeModules/compileTestAvx2.cpp

+
+#include <x86intrin.h>
+#include <xmmintrin.h> // SSE
+#include <emmintrin.h> // SSE2
+#include <pmmintrin.h> // SSE3
+#include <tmmintrin.h> // SSSE3
+#include <smmintrin.h> // SSE4
+
+#include <immintrin.h> // AVX
+
+int main(){
+	{
+	#ifdef __MIC__
+		__m512d res0d, res1d;
+		res0d = _mm512_hadd_pd(res0d, res1d);
+
+		__m512 res0, res1;
+		res0 = _mm512_hadd_ps(res0, res1);
+	#endif
+	}
+	{
+		__m256d res0d, res1d;
+		res0d = _mm256_hadd_pd(res0d, res1d);
+
+		__m256 res0, res1;
+		res0 = _mm256_hadd_ps(res0, res1);
+	}
+	{
+		__m128d res0d, res1d;
+		res0d = _mm_hadd_pd(res0d, res1d);
+
+		__m128 res0, res1;
+		res0 = _mm_hadd_ps(res0, res1);
+	}
+	return 0;
+}
+

Src/Utils/FAvx2.hpp

+// ===================================================================================
+// Copyright ScalFmm 2011 INRIA, Olivier Coulaud, Bérenger Bramas, Matthias Messner
+// olivier.coulaud@inria.fr, berenger.bramas@inria.fr
+// This software is a computer program whose purpose is to compute the FMM.
+//
+// This software is governed by the CeCILL-C and LGPL licenses and
+// abiding by the rules of distribution of free software.
+//
+// This program is distributed in the hope that it will be useful,
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+// GNU General Public and CeCILL-C Licenses for more details.
+// "http://www.cecill.info".
+// "http://www.gnu.org/licenses".
+// ===================================================================================
+#ifndef FAVX2_HPP
+#define FAVX2_HPP
+
+#include "FGlobal.hpp"
+#ifndef SCALFMM_USE_AVX2
+#error The AVX header is included while SCALFMM_USE_AVX is turned OFF
+#endif
+
+#include "immintrin.h"
+
+#ifdef __MIC__
+
+//Side effect operators DOUBLE
+inline __m512d& operator+=(__m512d & a, const __m512d & b){
+  return (a = _mm512_add_pd (a,b));
+}
+
+inline __m512d& operator-=(__m512d& a, const __m512d& b){
+  return (a = _mm512_sub_pd (a,b));
+}
+
+inline __m512d& operator*=(__m512d& a, const __m512d& b){
+  return (a = _mm512_mul_pd (a,b));
+}
+
+inline __m512d& operator/=(__m512d& a, const __m512d& b){
+  return (a = _mm512_div_pd (a,b));
+}
+
+//No side effect operators DOUBLE
+inline __m512d operator+(const __m512d& a,const  __m512d& b){
+  return _mm512_add_pd (a,b);
+}
+
+inline __m512d operator-(const __m512d& a, const __m512d& b){
+  return _mm512_sub_pd (a,b);
+}
+
+inline __m512d operator*(const __m512d& v1, const __m512d& v2){
+    return _mm512_mul_pd(v1, v2);
+}
+
+inline __m512d operator/(const __m512d& v1, const __m512d& v2){
+    return _mm512_div_pd(v1, v2);
+}
+
+//Side effect operators SINGLE
+inline __m512& operator+=(__m512 & a, const __m512 & b){
+  return (a = _mm512_add_ps (a,b));
+}
+
+inline __m512& operator-=(__m512& a, const __m512& b){
+  return (a = _mm512_sub_ps (a,b));
+}
+
+inline __m512& operator*=(__m512& a, const __m512& b){
+  return (a = _mm512_mul_ps (a,b));
+}
+
+inline __m512& operator/=(__m512& a, const __m512& b){
+  return (a = _mm512_div_ps (a,b));
+}
+
+//No side effect operators SINGLE
+inline __m512 operator+(const __m512& a,const  __m512& b){
+  return _mm512_add_ps (a,b);
+}
+
+inline __m512 operator-(const __m512& a, const __m512& b){
+  return _mm512_sub_ps (a,b);
+}
+
+inline __m512 operator*(const __m512& v1, const __m512& v2){
+    return _mm512_mul_ps(v1, v2);
+}
+
+inline __m512 operator/(const __m512& v1, const __m512& v2){
+    return _mm512_div_ps(v1, v2);
+}
+
+#endif
+
+#endif
+