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
753     A6 = _mm256_loadu_pd(W + 18);  in PetscKernel_A_gets_A_times_B_9()
759     S0 = _mm256_fmadd_pd(A6, B0, S0);  in PetscKernel_A_gets_A_times_B_9()
762     S3 = _mm256_fmadd_pd(A6, B1, S3);  in PetscKernel_A_gets_A_times_B_9()
765     S6 = _mm256_fmadd_pd(A6, B2, S6);  in PetscKernel_A_gets_A_times_B_9()
833     A6 = _mm256_loadu_pd(W + 63);  in PetscKernel_A_gets_A_times_B_9()
839     S0 = _mm256_fmadd_pd(A6, B6, S0);  in PetscKernel_A_gets_A_times_B_9()
842     S3 = _mm256_fmadd_pd(A6, B7, S3);  in PetscKernel_A_gets_A_times_B_9()
845     S6 = _mm256_fmadd_pd(A6, B8, S6);  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
989     A6 = _mm256_loadu_pd(A + 18);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1005     A6 = _mm256_fnmadd_pd(B0, C2, A6);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1021     A6 = _mm256_fnmadd_pd(B3, C5, A6);  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()
1053     A6 = _mm256_fnmadd_pd(B0, C2, A6);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1069     A6 = _mm256_fnmadd_pd(B3, C5, A6);  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()
1101     A6 = _mm256_fnmadd_pd(B0, C2, A6);  in PetscKernel_A_gets_A_minus_B_times_C_9()
1117     A6 = _mm256_fnmadd_pd(B3, C5, A6);  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()
1143     _mm256_storeu_pd(&A[18], A6);  in PetscKernel_A_gets_A_minus_B_times_C_9()