Merge branch 'Skylake-master' into master_pre_stage

configure.ac

@@ -853,12 +853,12 @@ if test x"${need_avx512}" = x"yes"; then
   AC_DEFINE([HAVE_AVX512],[1],[AVX512 is supported on this CPU])
 
   if test x"$can_compile_avx512" = x"yes"; then
-    AC_MSG_CHECKING([whether we compile for Xeon PHI or Xeon])
-    AC_COMPILE_IFELSE([AC_LANG_SOURCE([
+    AC_MSG_CHECKING([whether we compile for Xeon])
+    AC_RUN_IFELSE([AC_LANG_SOURCE([
      #include <x86intrin.h>
      int main(int argc, char **argv){
      __m512d sign;
-     __m512d h1;
+     __m512d h1_real;
 
      __m512d x1 = _mm512_xor_pd(h1_real, sign);
      return 0;
@@ -867,19 +867,37 @@ if test x"${need_avx512}" = x"yes"; then
      [can_compile_avx512_xeon=yes],
      [can_compile_avx512_xeon=no]
     )
-    AC_COMPILE_IFELSE([AC_LANG_SOURCE([
+    AC_MSG_RESULT([${can_compile_avx512_xeon}])
+
+    AC_MSG_CHECKING([whether we compile for Xeon PHI])
+    AC_RUN_IFELSE([AC_LANG_SOURCE([
      #include <x86intrin.h>
      int main(int argc, char **argv){
      __m512d sign;
      __m512d h1;
+     __m512d h2_real;
 
-     __m512d x1 = (__mm512d) _mm512_xor_epi64((__m512i) h2_real, (__m512i) sign);
+     __m512d x1 = (__m512d) _mm512_xor_epi64((__m512i) h2_real, (__m512i) sign);
      return 0;
      }
      ])],
      [can_compile_avx512_xeon_phi=yes],
      [can_compile_avx512_xeon_phi=no]
     )
+    AC_MSG_RESULT([${can_compile_avx512_xeon_phi}])
+
+    # this is needed for the intel compiler
+    if test x"$can_compile_avx512_xeon" = x"yes" ; then
+      if test x"$can_compile_avx512_xeon_phi" = x"yes" ; then
+        # we want only one to be true; this is ugly but could not come up with a better way
+        grep Phi /proc/cpuinfo > /dev/null
+        if test x"$?" = x"0" ; then
+	  echo "Xeon PHI found ... disabling AVX512 Xeon"
+          can_compile_avx512_xeon=no
+        fi
+      fi
+    fi
+
     if test x"$can_compile_avx512_xeon" = x"yes"; then
       AC_DEFINE([HAVE_AVX512_XEON],[1],[AVX512 for Xeon is supported on this CPU])
     else

src/elpa2/kernels/complex_avx512_2hv_template.c

-XEON_PHI/    This file is part of ELPA.
+//    This file is part of ELPA.
 //
 //    The ELPA library was originally created by the ELPA consortium,
 //    consisting of the following organizations:

src/elpa2/kernels/real_avx512_2hv_template.c

@@ -345,6 +345,7 @@ void double_hh_trafo_real_avx512_2hv_single(float* q, float* hh, int* pnb, int*
 #ifdef SINGLE_PRECISION_REAL
         h1 = (__AVX512_DATATYPE) _mm512_xor_epi32((__AVX512i) tau1, (__AVX512i) sign);
 #endif
+#endif
 
 #ifdef HAVE_AVX512_XEON
 #if defined(DOUBLE_PRECISION_REAL) || defined(SINGLE_PRECISION_REAL)
@@ -364,6 +365,7 @@ void double_hh_trafo_real_avx512_2hv_single(float* q, float* hh, int* pnb, int*
 	h1 = (__AVX512_DATATYPE) _mm512_xor_epi32((__AVX512i) tau2, (__AVX512i) sign);
 #endif
 #endif
+
 #ifdef HAVE_AVX512_XEON
 #if defined(DOUBLE_PRECISION_REAL) || defined(SINGLE_PRECISION_REAL)
 	h1 = _AVX512_XOR(tau2, sign);
@@ -441,7 +443,6 @@ void double_hh_trafo_real_avx512_2hv_single(float* q, float* hh, int* pnb, int*
 	q4 = _AVX512_LOAD(&q[(nb*ldq)+3*offset]);
 	q4 = _AVX512_FMA(x4, h1, q4);
 	_AVX512_STORE(&q[(nb*ldq)+3*offset],q4);
->>>>>>> Skylake
 
 }
 
@@ -531,6 +532,7 @@ void double_hh_trafo_real_avx512_2hv_single(float* q, float* hh, int* pnb, int*
         h1 = (__AVX512_DATATYPE) _mm512_xor_epi32((__AVX512i) tau1, (__AVX512i) sign);
 #endif
 #endif
+
 #ifdef HAVE_AVX512_XEON
 #if defined(DOUBLE_PRECISION_REAL) || defined(SINGLE_PRECISION_REAL)
 	h1 = _AVX512_XOR(tau1, sign);
@@ -549,6 +551,7 @@ void double_hh_trafo_real_avx512_2hv_single(float* q, float* hh, int* pnb, int*
 	h1 = (__AVX512_DATATYPE) _mm512_xor_epi32((__AVX512i) tau2, (__AVX512i) sign);
 #endif
 #endif
+
 #ifdef HAVE_AVX512_XEON
 #if defined(DOUBLE_PRECISION_REAL) || defined(SINGLE_PRECISION_REAL)
 	h1 = _AVX512_XOR(tau2, sign);
@@ -613,7 +616,6 @@ void double_hh_trafo_real_avx512_2hv_single(float* q, float* hh, int* pnb, int*
 	q3 = _AVX512_LOAD(&q[(nb*ldq)+2*offset]);
 	q3 = _AVX512_FMA(x3, h1, q3);
 	_AVX512_STORE(&q[(nb*ldq)+2*offset],q3);
->>>>>>> Skylake
 
 }
 
@@ -693,6 +695,7 @@ void double_hh_trafo_real_avx512_2hv_single(float* q, float* hh, int* pnb, int*
         h1 = (__AVX512_DATATYPE) _mm512_xor_epi32((__AVX512i) tau1, (__AVX512i) sign);
 #endif
 #endif
+
 #ifdef HAVE_AVX512_XEON
 #if defined(DOUBLE_PRECISION_REAL) || defined(SINGLE_PRECISION_REAL)
 	h1 = _AVX512_XOR(tau1, sign);
@@ -708,6 +711,7 @@ void double_hh_trafo_real_avx512_2hv_single(float* q, float* hh, int* pnb, int*
 	h1 = (__AVX512_DATATYPE) _mm512_xor_epi32((__AVX512i) tau2, (__AVX512i) sign);
 #endif
 #endif
+
 #ifdef HAVE_AVX512_XEON
 #if defined(DOUBLE_PRECISION_REAL) || defined(SINGLE_PRECISION_REAL)
 	h1 = _AVX512_XOR(tau2, sign);
@@ -758,8 +762,6 @@ void double_hh_trafo_real_avx512_2hv_single(float* q, float* hh, int* pnb, int*
 	q2 = _AVX512_LOAD(&q[(nb*ldq)+offset]);
 	q2 = _AVX512_FMA(x2, h1, q2);
 	_AVX512_STORE(&q[(nb*ldq)+offset],q2);
->>>>>>> Skylake
-
 }
 
 /**
@@ -830,6 +832,7 @@ void double_hh_trafo_real_avx512_2hv_single(float* q, float* hh, int* pnb, int*
         h1 = (__AVX512_DATATYPE) _mm512_xor_epi32((__AVX512i) tau1, (__AVX512i) sign);
 #endif
 #endif
+
 #ifdef HAVE_AVX512_XEON
 #if defined(DOUBLE_PRECISION_REAL) || defined(SINGLE_PRECISION_REAL)
 	h1 = _AVX512_XOR(tau1, sign);
@@ -837,6 +840,7 @@ void double_hh_trafo_real_avx512_2hv_single(float* q, float* hh, int* pnb, int*
 #endif
 
 	x1 = _AVX512_MUL(x1, h1);
+
 #ifdef HAVE_AVX512_XEON_PHI
 #ifdef DOUBLE_PRECISION_REAL
         h1 = (__AVX512_DATATYPE) _mm512_xor_epi64((__AVX512i) tau2, (__AVX512i) sign);
@@ -845,6 +849,7 @@ void double_hh_trafo_real_avx512_2hv_single(float* q, float* hh, int* pnb, int*
         h1 = (__AVX512_DATATYPE) _mm512_xor_epi32((__AVX512i) tau2, (__AVX512i) sign);
 #endif
 #endif
+
 #ifdef HAVE_AVX512_XEON
 #if defined(DOUBLE_PRECISION_REAL) || defined(SINGLE_PRECISION_REAL)
 	h1 = _AVX512_XOR(tau2, sign);