707 …6d  S0, S1, S2, S3, S4, S5, S6, S7, S8, B0, B1, B2, B6, B7, B8, A0, A1, A2, A3, A4, A5, A6, A7, A8;  in PetscKernel_A_gets_A_times_B_9()  local
754     A7 = _mm256_loadu_pd(W + 22);  in PetscKernel_A_gets_A_times_B_9()
760     S1 = _mm256_fmadd_pd(A7, B0, S1);  in PetscKernel_A_gets_A_times_B_9()
763     S4 = _mm256_fmadd_pd(A7, B1, S4);  in PetscKernel_A_gets_A_times_B_9()
766     S7 = _mm256_fmadd_pd(A7, B2, S7);  in PetscKernel_A_gets_A_times_B_9()
834     A7 = _mm256_loadu_pd(W + 67);  in PetscKernel_A_gets_A_times_B_9()
840     S1 = _mm256_fmadd_pd(A7, B6, S1);  in PetscKernel_A_gets_A_times_B_9()
843     S4 = _mm256_fmadd_pd(A7, B7, S4);  in PetscKernel_A_gets_A_times_B_9()
846     S7 = _mm256_fmadd_pd(A7, B8, 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
990     A7 = _mm256_loadu_pd(A + 22);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1006     A7 = _mm256_fnmadd_pd(B1, C2, A7);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1022     A7 = _mm256_fnmadd_pd(B4, C5, A7);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1038     A7 = _mm256_fnmadd_pd(B7, C8, A7);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1054     A7 = _mm256_fnmadd_pd(B1, C2, A7);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1070     A7 = _mm256_fnmadd_pd(B4, C5, A7);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1086     A7 = _mm256_fnmadd_pd(B7, C8, A7);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1102     A7 = _mm256_fnmadd_pd(B1, C2, A7);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1118     A7 = _mm256_fnmadd_pd(B4, C5, A7);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1134     A7 = _mm256_fnmadd_pd(B7, C8, A7);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1144     _mm256_storeu_pd(&A[22], A7);  in PetscKernel_A_gets_A_minus_B_times_C_9()