Lines Matching refs:A3
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
737 A3 = _mm256_loadu_pd(W + 9); in PetscKernel_A_gets_A_times_B_9()
743 S0 = _mm256_fmadd_pd(A3, B6, S0); in PetscKernel_A_gets_A_times_B_9()
746 S3 = _mm256_fmadd_pd(A3, B7, S3); in PetscKernel_A_gets_A_times_B_9()
749 S6 = _mm256_fmadd_pd(A3, B8, S6); in PetscKernel_A_gets_A_times_B_9()
785 A3 = _mm256_loadu_pd(W + 36); in PetscKernel_A_gets_A_times_B_9()
791 S0 = _mm256_fmadd_pd(A3, B0, S0); in PetscKernel_A_gets_A_times_B_9()
794 S3 = _mm256_fmadd_pd(A3, B1, S3); in PetscKernel_A_gets_A_times_B_9()
797 S6 = _mm256_fmadd_pd(A3, B2, S6); in PetscKernel_A_gets_A_times_B_9()
817 A3 = _mm256_loadu_pd(W + 54); in PetscKernel_A_gets_A_times_B_9()
823 S0 = _mm256_fmadd_pd(A3, B0, S0); in PetscKernel_A_gets_A_times_B_9()
826 S3 = _mm256_fmadd_pd(A3, B1, S3); in PetscKernel_A_gets_A_times_B_9()
829 S6 = _mm256_fmadd_pd(A3, 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
986 A3 = _mm256_loadu_pd(A + 9); in PetscKernel_A_gets_A_minus_B_times_C_9()
1002 A3 = _mm256_fnmadd_pd(B0, C1, A3); in PetscKernel_A_gets_A_minus_B_times_C_9()
1018 A3 = _mm256_fnmadd_pd(B3, C4, A3); 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()
1050 A3 = _mm256_fnmadd_pd(B0, C1, A3); in PetscKernel_A_gets_A_minus_B_times_C_9()
1066 A3 = _mm256_fnmadd_pd(B3, C4, A3); 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()
1098 A3 = _mm256_fnmadd_pd(B0, C1, A3); in PetscKernel_A_gets_A_minus_B_times_C_9()
1114 A3 = _mm256_fnmadd_pd(B3, C4, A3); 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()
1140 _mm256_storeu_pd(&A[9], A3); in PetscKernel_A_gets_A_minus_B_times_C_9()