Use fmaddsub

code/matrix_multiplication.h

@@ -107,9 +107,8 @@ void matrixMultiplyAVX2(const std::complex<float>* a,
   __m256 a_4_inv = _mm256_mul_ps(a_4, inv);
   __m256 b_4 = _mm256_permutevar8x32_ps(b_m, b_4_ind);
 
-  __m256 c_m = _mm256_mul_ps(a_1, b_1);
-  c_m = _mm256_fmadd_ps(a_2, b_2, c_m);
-  c_m = _mm256_fmadd_ps(a_3_inv, b_3, c_m);
-  c_m = _mm256_fmadd_ps(a_4_inv, b_4, c_m);
+  __m256 c_p1 = _mm256_fmaddsub_ps(a_1, b_1, _mm256_mul_ps(a_3, b_3));
+  __m256 c_p2 = _mm256_fmaddsub_ps(a_2, b_2, _mm256_mul_ps(a_4, b_4));
+  __m256 c_m = _mm256_add_ps(c_p1, c_p2);
   _mm256_store_ps(c_ptr, c_m);
 }
\ No newline at end of file