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
741     B7 = _mm256_broadcast_sd(B + 10);  in PetscKernel_A_gets_A_times_B_9()
746     S3 = _mm256_fmadd_pd(A3, B7, S3);  in PetscKernel_A_gets_A_times_B_9()
747     S4 = _mm256_fmadd_pd(A4, B7, S4);  in PetscKernel_A_gets_A_times_B_9()
748     S5 = _mm256_fmadd_pd(A5, B7, S5);  in PetscKernel_A_gets_A_times_B_9()
773     B7 = _mm256_broadcast_sd(B + 12);  in PetscKernel_A_gets_A_times_B_9()
778     S3 = _mm256_fmadd_pd(A0, B7, S3);  in PetscKernel_A_gets_A_times_B_9()
779     S4 = _mm256_fmadd_pd(A1, B7, S4);  in PetscKernel_A_gets_A_times_B_9()
780     S5 = _mm256_fmadd_pd(A2, B7, S5);  in PetscKernel_A_gets_A_times_B_9()
805     B7 = _mm256_broadcast_sd(B + 14);  in PetscKernel_A_gets_A_times_B_9()
810     S3 = _mm256_fmadd_pd(A0, B7, S3);  in PetscKernel_A_gets_A_times_B_9()
811     S4 = _mm256_fmadd_pd(A1, B7, S4);  in PetscKernel_A_gets_A_times_B_9()
812     S5 = _mm256_fmadd_pd(A2, B7, S5);  in PetscKernel_A_gets_A_times_B_9()
837     B7 = _mm256_broadcast_sd(B + 16);  in PetscKernel_A_gets_A_times_B_9()
842     S3 = _mm256_fmadd_pd(A6, B7, S3);  in PetscKernel_A_gets_A_times_B_9()
843     S4 = _mm256_fmadd_pd(A7, B7, S4);  in PetscKernel_A_gets_A_times_B_9()
844     S5 = _mm256_fmadd_pd(A8, B7, S5);  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
1026     B7 = _mm256_loadu_pd(B + 22);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1032     A1 = _mm256_fnmadd_pd(B7, C6, A1);  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()
1038     A7 = _mm256_fnmadd_pd(B7, C8, A7);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1074     B7 = _mm256_loadu_pd(B + 49);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1080     A1 = _mm256_fnmadd_pd(B7, C6, A1);  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()
1086     A7 = _mm256_fnmadd_pd(B7, C8, A7);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1122     B7 = _mm256_loadu_pd(B + 76);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1128     A1 = _mm256_fnmadd_pd(B7, C6, A1);  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()
1134     A7 = _mm256_fnmadd_pd(B7, C8, A7);  in PetscKernel_A_gets_A_minus_B_times_C_9()