707 …__m256d  S0, S1, S2, S3, S4, S5, S6, S7, S8, B0, B1, B2, B6, B7, B8, A0, A1, A2, A3, A4, A5, A6, A…  in PetscKernel_A_gets_A_times_B_9()  local
738     A4 = _mm256_loadu_pd(W + 13);  in PetscKernel_A_gets_A_times_B_9()
744     S1 = _mm256_fmadd_pd(A4, B6, S1);  in PetscKernel_A_gets_A_times_B_9()
747     S4 = _mm256_fmadd_pd(A4, B7, S4);  in PetscKernel_A_gets_A_times_B_9()
750     S7 = _mm256_fmadd_pd(A4, B8, S7);  in PetscKernel_A_gets_A_times_B_9()
786     A4 = _mm256_loadu_pd(W + 40);  in PetscKernel_A_gets_A_times_B_9()
792     S1 = _mm256_fmadd_pd(A4, B0, S1);  in PetscKernel_A_gets_A_times_B_9()
795     S4 = _mm256_fmadd_pd(A4, B1, S4);  in PetscKernel_A_gets_A_times_B_9()
798     S7 = _mm256_fmadd_pd(A4, B2, S7);  in PetscKernel_A_gets_A_times_B_9()
818     A4 = _mm256_loadu_pd(W + 58);  in PetscKernel_A_gets_A_times_B_9()
824     S1 = _mm256_fmadd_pd(A4, B0, S1);  in PetscKernel_A_gets_A_times_B_9()
827     S4 = _mm256_fmadd_pd(A4, B1, S4);  in PetscKernel_A_gets_A_times_B_9()
830     S7 = _mm256_fmadd_pd(A4, B2, S7);  in PetscKernel_A_gets_A_times_B_9()
980 …__m256d  A0, A1, A2, A3, A4, A5, A6, A7, A8, B0, B1, B2, B3, B4, B5, B6, B7, B8, C0, C1, C2, C3, C…  in PetscKernel_A_gets_A_minus_B_times_C_9()  local
987     A4 = _mm256_loadu_pd(A + 13);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1003     A4 = _mm256_fnmadd_pd(B1, C1, A4);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1019     A4 = _mm256_fnmadd_pd(B4, C4, A4);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1035     A4 = _mm256_fnmadd_pd(B7, C7, A4);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1051     A4 = _mm256_fnmadd_pd(B1, C1, A4);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1067     A4 = _mm256_fnmadd_pd(B4, C4, A4);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1083     A4 = _mm256_fnmadd_pd(B7, C7, A4);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1099     A4 = _mm256_fnmadd_pd(B1, C1, A4);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1115     A4 = _mm256_fnmadd_pd(B4, C4, A4);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1131     A4 = _mm256_fnmadd_pd(B7, C7, A4);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1141     _mm256_storeu_pd(&A[13], A4);  in PetscKernel_A_gets_A_minus_B_times_C_9()