Fix small AVX512 alignment issue

Exchange _mm512_load_pd()/_mm512_load_ps() of stack variables with
_mm512_set_pd, as the stack variables might not have proper alignment!

src/elpa2/kernels/elpa2_kernels_complex_avx512_2hv_double_precision.c

@@ -114,16 +114,6 @@ static __forceinline void hh_trafo_complex_kernel_16_AVX512_2hv_double(double co
 	double* q_dbl = (double*)q;
 	double* hh_dbl = (double*)hh;
 	double* s_dbl = (double*)(&s);
-        double s_helper[8];
-
-        s_helper[0] = s_dbl[0];
-        s_helper[1] = s_dbl[1];
-        s_helper[2] = s_dbl[0];
-        s_helper[3] = s_dbl[1];
-        s_helper[4] = s_dbl[0];
-        s_helper[5] = s_dbl[1];
-        s_helper[6] = s_dbl[0];
-        s_helper[7] = s_dbl[1];
 
 	__m512d x1, x2, x3, x4;
 	__m512d y1, y2, y3, y4;
@@ -281,7 +271,10 @@ static __forceinline void hh_trafo_complex_kernel_16_AVX512_2hv_double(double co
 //        tmp2 = _mm512_broadcast_f64x2(_mm512_castpd512_pd128(tmp_s));
 //        tmp2 = _mm512_set4_pd(s_dbl[0],s_dbl[1], s_dbl[0],s_dbl[1]);
 
-	tmp2 = _mm512_load_pd(&s_helper);
+	tmp2 = _mm512_set_pd(s_dbl[1], s_dbl[0],
+			     s_dbl[1], s_dbl[0],
+			     s_dbl[1], s_dbl[0],
+			     s_dbl[1], s_dbl[0]);
 	tmp1 = _mm512_mul_pd(h2_imag, tmp2);
 
 	tmp2 = _mm512_FMADDSUB_pd(h2_real, tmp2, _mm512_shuffle_pd(tmp1, tmp1, 0x55));
@@ -465,16 +458,6 @@ static __forceinline void hh_trafo_complex_kernel_8_AVX512_2hv_double(double com
 	double* q_dbl = (double*)q;
 	double* hh_dbl = (double*)hh;
 	double* s_dbl = (double*)(&s);
-        double s_helper[8];
-
-        s_helper[0] = s_dbl[0];
-        s_helper[1] = s_dbl[1];
-        s_helper[2] = s_dbl[0];
-        s_helper[3] = s_dbl[1];
-        s_helper[4] = s_dbl[0];
-        s_helper[5] = s_dbl[1];
-        s_helper[6] = s_dbl[0];
-        s_helper[7] = s_dbl[1];
 
 	__m512d x1, x2;
 	__m512d y1, y2;
@@ -584,8 +567,10 @@ static __forceinline void hh_trafo_complex_kernel_8_AVX512_2hv_double(double com
 //       tmp2 = _mm512_broadcast_f64x2(_mm512_castpd512_pd128(tmp_s));
  //       tmp2 = _mm512_set4_pd(s_dbl[0],s_dbl[1], s_dbl[0],s_dbl[1]);
 
-        tmp2 = _mm512_load_pd(&s_helper);
-
+	tmp2 = _mm512_set_pd(s_dbl[1], s_dbl[0],
+			     s_dbl[1], s_dbl[0],
+			     s_dbl[1], s_dbl[0],
+			     s_dbl[1], s_dbl[0]);
 	tmp1 = _mm512_mul_pd(h2_imag, tmp2);
 
 	tmp2 = _mm512_FMADDSUB_pd(h2_real, tmp2, _mm512_shuffle_pd(tmp1, tmp1, 0x55));

src/elpa2/kernels/elpa2_kernels_complex_avx512_2hv_single_precision.c

@@ -112,25 +112,6 @@ static __forceinline void hh_trafo_complex_kernel_32_AVX512_2hv_single(float com
 	float* q_dbl = (float*)q;
 	float* hh_dbl = (float*)hh;
 	float* s_dbl = (float*)(&s);
-        float s_helper[16];
-
-        s_helper[0] = s_dbl[0];
-        s_helper[1] = s_dbl[1];
-        s_helper[2] = s_dbl[0];
-        s_helper[3] = s_dbl[1];
-        s_helper[4] = s_dbl[0];
-        s_helper[5] = s_dbl[1];
-        s_helper[6] = s_dbl[0];
-        s_helper[7] = s_dbl[1];
-        s_helper[8] = s_dbl[0];
-        s_helper[9] = s_dbl[1];
-        s_helper[10] = s_dbl[0];
-        s_helper[11] = s_dbl[1];
-        s_helper[12] = s_dbl[0];
-        s_helper[13] = s_dbl[1];
-        s_helper[14] = s_dbl[0];
-        s_helper[15] = s_dbl[1];
-
 
 	__m512 x1, x2, x3, x4;
 	__m512 y1, y2, y3, y4;
@@ -287,7 +268,7 @@ static __forceinline void hh_trafo_complex_kernel_32_AVX512_2hv_single(float com
 //        tmp2 = _mm512_broadcast_f64x2(_mm512_castpd512_ps128(tmp_s));
 //        tmp2 = _mm512_set4_ps(s_dbl[0],s_dbl[1], s_dbl[0],s_dbl[1]);
 
-	tmp2 = _mm512_load_ps(&s_helper);
+	tmp2 = (__m512) _mm512_set1_pd(*(double*)(&s_dbl[0]));
 	tmp1 = _mm512_mul_ps(h2_imag, tmp2);
 
 	tmp2 = _mm512_FMADDSUB_ps(h2_real, tmp2, _mm512_shuffle_ps(tmp1, tmp1, 0xb1));
@@ -465,24 +446,7 @@ static __forceinline void hh_trafo_complex_kernel_16_AVX512_2hv_single(float com
 	float* q_dbl = (float*)q;
 	float* hh_dbl = (float*)hh;
 	float* s_dbl = (float*)(&s);
-	float s_helper[16];
-
-        s_helper[0] = s_dbl[0];
-        s_helper[1] = s_dbl[1];
-        s_helper[2] = s_dbl[0];
-        s_helper[3] = s_dbl[1];
-        s_helper[4] = s_dbl[0];
-        s_helper[5] = s_dbl[1];
-        s_helper[6] = s_dbl[0];
-        s_helper[7] = s_dbl[1];
-        s_helper[8] = s_dbl[0];
-        s_helper[9] = s_dbl[1];
-        s_helper[10] = s_dbl[0];
-        s_helper[11] = s_dbl[1];
-        s_helper[12] = s_dbl[0];
-        s_helper[13] = s_dbl[1];
-        s_helper[14] = s_dbl[0];
-        s_helper[15] = s_dbl[1];
+
 	__m512 x1, x2;
 	__m512 y1, y2;
 	__m512 q1, q2;
@@ -590,8 +554,7 @@ static __forceinline void hh_trafo_complex_kernel_16_AVX512_2hv_single(float com
 //       tmp2 = _mm512_broadcast_f64x2(_mm512_castpd512_ps128(tmp_s));
 //        tmp2 = _mm512_set4_ps(s_dbl[0],s_dbl[1], s_dbl[0],s_dbl[1]);
 
-	tmp2 = _mm512_load_ps(&s_helper);
-
+	tmp2 = (__m512) _mm512_set1_pd(*(double*)(&s_dbl[0]));
 	tmp1 = _mm512_mul_ps(h2_imag, tmp2);
 
 	tmp2 = _mm512_FMADDSUB_ps(h2_real, tmp2, _mm512_shuffle_ps(tmp1, tmp1, 0xb1));