313 #define DEF_VMAE(BITS, TBITS)                                                  \  argument
319     for (i = 0, j = 0; i < (128 / TBITS); i++, j += 2) {                       \
320         int##TBITS##_t a = (int##BITS##_t)s390_vec_read_element##BITS(v2, j);  \
321         int##TBITS##_t b = (int##BITS##_t)s390_vec_read_element##BITS(v3, j);  \
322         int##TBITS##_t c = s390_vec_read_element##TBITS(v4, i);                \
324         s390_vec_write_element##TBITS(v1, i, a * b + c);                       \
331 #define DEF_VMALE(BITS, TBITS)                                                 \  argument
337     for (i = 0, j = 0; i < (128 / TBITS); i++, j += 2) {                       \
338         uint##TBITS##_t a = s390_vec_read_element##BITS(v2, j);                \
339         uint##TBITS##_t b = s390_vec_read_element##BITS(v3, j);                \
340         uint##TBITS##_t c = s390_vec_read_element##TBITS(v4, i);               \
342         s390_vec_write_element##TBITS(v1, i, a * b + c);                       \
349 #define DEF_VMAO(BITS, TBITS)                                                  \  argument
355     for (i = 0, j = 1; i < (128 / TBITS); i++, j += 2) {                       \
356         int##TBITS##_t a = (int##BITS##_t)s390_vec_read_element##BITS(v2, j);  \
357         int##TBITS##_t b = (int##BITS##_t)s390_vec_read_element##BITS(v3, j);  \
358         int##TBITS##_t c = s390_vec_read_element##TBITS(v4, i);                \
360         s390_vec_write_element##TBITS(v1, i, a * b + c);                       \
367 #define DEF_VMALO(BITS, TBITS)                                                 \  argument
373     for (i = 0, j = 1; i < (128 / TBITS); i++, j += 2) {                       \
374         uint##TBITS##_t a = s390_vec_read_element##BITS(v2, j);                \
375         uint##TBITS##_t b = s390_vec_read_element##BITS(v3, j);                \
376         uint##TBITS##_t c = s390_vec_read_element##TBITS(v4, i);               \
378         s390_vec_write_element##TBITS(v1, i, a * b + c);                       \
417 #define DEF_VME(BITS, TBITS)                                                   \  argument
423     for (i = 0, j = 0; i < (128 / TBITS); i++, j += 2) {                       \
424         int##TBITS##_t a = (int##BITS##_t)s390_vec_read_element##BITS(v2, j);  \
425         int##TBITS##_t b = (int##BITS##_t)s390_vec_read_element##BITS(v3, j);  \
427         s390_vec_write_element##TBITS(v1, i, a * b);                           \
434 #define DEF_VMLE(BITS, TBITS)                                                  \  argument
440     for (i = 0, j = 0; i < (128 / TBITS); i++, j += 2) {                       \
441         const uint##TBITS##_t a = s390_vec_read_element##BITS(v2, j);          \
442         const uint##TBITS##_t b = s390_vec_read_element##BITS(v3, j);          \
444         s390_vec_write_element##TBITS(v1, i, a * b);                           \
451 #define DEF_VMO(BITS, TBITS)                                                   \  argument
457     for (i = 0, j = 1; i < (128 / TBITS); i++, j += 2) {                       \
458         int##TBITS##_t a = (int##BITS##_t)s390_vec_read_element##BITS(v2, j);  \
459         int##TBITS##_t b = (int##BITS##_t)s390_vec_read_element##BITS(v3, j);  \
461         s390_vec_write_element##TBITS(v1, i, a * b);                           \
468 #define DEF_VMLO(BITS, TBITS)                                                  \  argument
474     for (i = 0, j = 1; i < (128 / TBITS); i++, j += 2) {                       \
475         const uint##TBITS##_t a = s390_vec_read_element##BITS(v2, j);          \
476         const uint##TBITS##_t b = s390_vec_read_element##BITS(v3, j);          \
478         s390_vec_write_element##TBITS(v1, i, a * b);                           \