Lines Matching refs:A0

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
721 A0 = _mm256_loadu_pd(W + 0); in PetscKernel_A_gets_A_times_B_9()
727 S0 = _mm256_fmadd_pd(A0, B0, S0); in PetscKernel_A_gets_A_times_B_9()
730 S3 = _mm256_fmadd_pd(A0, B1, S3); in PetscKernel_A_gets_A_times_B_9()
733 S6 = _mm256_fmadd_pd(A0, B2, S6); in PetscKernel_A_gets_A_times_B_9()
769 A0 = _mm256_loadu_pd(W + 27); in PetscKernel_A_gets_A_times_B_9()
775 S0 = _mm256_fmadd_pd(A0, B6, S0); in PetscKernel_A_gets_A_times_B_9()
778 S3 = _mm256_fmadd_pd(A0, B7, S3); in PetscKernel_A_gets_A_times_B_9()
781 S6 = _mm256_fmadd_pd(A0, B8, S6); in PetscKernel_A_gets_A_times_B_9()
801 A0 = _mm256_loadu_pd(W + 45); in PetscKernel_A_gets_A_times_B_9()
807 S0 = _mm256_fmadd_pd(A0, B6, S0); in PetscKernel_A_gets_A_times_B_9()
810 S3 = _mm256_fmadd_pd(A0, B7, S3); in PetscKernel_A_gets_A_times_B_9()
813 S6 = _mm256_fmadd_pd(A0, B8, S6); in PetscKernel_A_gets_A_times_B_9()
849 A0 = _mm256_loadu_pd(W + 72); in PetscKernel_A_gets_A_times_B_9()
855 S0 = _mm256_fmadd_pd(A0, B0, S0); in PetscKernel_A_gets_A_times_B_9()
858 S3 = _mm256_fmadd_pd(A0, B1, S3); in PetscKernel_A_gets_A_times_B_9()
861 S6 = _mm256_fmadd_pd(A0, B2, 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
983 A0 = _mm256_loadu_pd(A + 0); in PetscKernel_A_gets_A_minus_B_times_C_9()
999 A0 = _mm256_fnmadd_pd(B0, C0, A0); in PetscKernel_A_gets_A_minus_B_times_C_9()
1015 A0 = _mm256_fnmadd_pd(B3, C3, A0); 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()
1047 A0 = _mm256_fnmadd_pd(B0, C0, A0); in PetscKernel_A_gets_A_minus_B_times_C_9()
1063 A0 = _mm256_fnmadd_pd(B3, C3, A0); 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()
1095 A0 = _mm256_fnmadd_pd(B0, C0, A0); in PetscKernel_A_gets_A_minus_B_times_C_9()
1111 A0 = _mm256_fnmadd_pd(B3, C3, A0); 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()
1137 _mm256_storeu_pd(&A[0], A0); in PetscKernel_A_gets_A_minus_B_times_C_9()