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
755     A8 = _mm256_loadu_pd(W + 26);  in PetscKernel_A_gets_A_times_B_9()
761     S2 = _mm256_fmadd_pd(A8, B0, S2);  in PetscKernel_A_gets_A_times_B_9()
764     S5 = _mm256_fmadd_pd(A8, B1, S5);  in PetscKernel_A_gets_A_times_B_9()
767     S8 = _mm256_fmadd_pd(A8, B2, S8);  in PetscKernel_A_gets_A_times_B_9()
835     A8 = _mm256_loadu_pd(W + 71);  in PetscKernel_A_gets_A_times_B_9()
841     S2 = _mm256_fmadd_pd(A8, B6, S2);  in PetscKernel_A_gets_A_times_B_9()
844     S5 = _mm256_fmadd_pd(A8, B7, S5);  in PetscKernel_A_gets_A_times_B_9()
847     S8 = _mm256_fmadd_pd(A8, B8, S8);  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
991     A8 = _mm256_loadu_pd(A + 26);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1007     A8 = _mm256_fnmadd_pd(B2, C2, A8);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1023     A8 = _mm256_fnmadd_pd(B5, C5, A8);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1039     A8 = _mm256_fnmadd_pd(B8, C8, A8);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1055     A8 = _mm256_fnmadd_pd(B2, C2, A8);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1071     A8 = _mm256_fnmadd_pd(B5, C5, A8);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1087     A8 = _mm256_fnmadd_pd(B8, C8, A8);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1103     A8 = _mm256_fnmadd_pd(B2, C2, A8);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1119     A8 = _mm256_fnmadd_pd(B5, C5, A8);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1135     A8 = _mm256_fnmadd_pd(B8, C8, A8);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1145     _mm256_maskstore_pd(&A[26], _mm256_set_epi64x(0LL, 0LL, 0LL, 1LL << 63), A8);  in PetscKernel_A_gets_A_minus_B_times_C_9()