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
740     B6 = _mm256_broadcast_sd(B + 1);  in PetscKernel_A_gets_A_times_B_9()
743     S0 = _mm256_fmadd_pd(A3, B6, S0);  in PetscKernel_A_gets_A_times_B_9()
744     S1 = _mm256_fmadd_pd(A4, B6, S1);  in PetscKernel_A_gets_A_times_B_9()
745     S2 = _mm256_fmadd_pd(A5, B6, S2);  in PetscKernel_A_gets_A_times_B_9()
772     B6 = _mm256_broadcast_sd(B + 3);  in PetscKernel_A_gets_A_times_B_9()
775     S0 = _mm256_fmadd_pd(A0, B6, S0);  in PetscKernel_A_gets_A_times_B_9()
776     S1 = _mm256_fmadd_pd(A1, B6, S1);  in PetscKernel_A_gets_A_times_B_9()
777     S2 = _mm256_fmadd_pd(A2, B6, S2);  in PetscKernel_A_gets_A_times_B_9()
804     B6 = _mm256_broadcast_sd(B + 5);  in PetscKernel_A_gets_A_times_B_9()
807     S0 = _mm256_fmadd_pd(A0, B6, S0);  in PetscKernel_A_gets_A_times_B_9()
808     S1 = _mm256_fmadd_pd(A1, B6, S1);  in PetscKernel_A_gets_A_times_B_9()
809     S2 = _mm256_fmadd_pd(A2, B6, S2);  in PetscKernel_A_gets_A_times_B_9()
836     B6 = _mm256_broadcast_sd(B + 7);  in PetscKernel_A_gets_A_times_B_9()
839     S0 = _mm256_fmadd_pd(A6, B6, S0);  in PetscKernel_A_gets_A_times_B_9()
840     S1 = _mm256_fmadd_pd(A7, B6, S1);  in PetscKernel_A_gets_A_times_B_9()
841     S2 = _mm256_fmadd_pd(A8, B6, S2);  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
1025     B6 = _mm256_loadu_pd(B + 18);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1031     A0 = _mm256_fnmadd_pd(B6, C6, A0);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1034     A3 = _mm256_fnmadd_pd(B6, C7, A3);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1037     A6 = _mm256_fnmadd_pd(B6, C8, A6);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1073     B6 = _mm256_loadu_pd(B + 45);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1079     A0 = _mm256_fnmadd_pd(B6, C6, A0);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1082     A3 = _mm256_fnmadd_pd(B6, C7, A3);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1085     A6 = _mm256_fnmadd_pd(B6, C8, A6);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1121     B6 = _mm256_loadu_pd(B + 72);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1127     A0 = _mm256_fnmadd_pd(B6, C6, A0);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1130     A3 = _mm256_fnmadd_pd(B6, C7, A3);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1133     A6 = _mm256_fnmadd_pd(B6, C8, A6);  in PetscKernel_A_gets_A_minus_B_times_C_9()