target/arm: Split zdn in do_z2z_n1 and do_z2z_n1_fpst
Rename from z2z to zzz to indicate zd and zn may not overlap. During decode, retain z2z name and duplicate the same field into zd and zn. Reviewed-by: Peter Maydell <peter.maydell@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20260826174213.614571-2-richard.henderson@linaro.org Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Richard Henderson committed
Aug 26, 2026 at 10:41 UTC
4689fe9858dab51b7dc0d94fa3df8a3676dd8b60
2 files changed
+39
-40
target/arm/tcg/sme.decode
+4
-4
@@ -220,11 +220,13 @@ UMOPA2_s 1010000 1 10 0 ..... ... ... ..... . 10 .. @op_32
220
%zd_ax2 1:4 !function=times_2
221
%zd_ax4 2:3 !function=times_4
222
223
+&zzz_en zd zn zm esz n
224
&z2z_en zdn zm esz n
225
+
226
@z2z_2x1 ....... . esz:2 .. zm:4 ....0. ..... .... . \
225
- &z2z_en n=2 zdn=%zd_ax2
227
+ &zzz_en n=2 zd=%zd_ax2 zn=%zd_ax2
228
@z2z_4x1 ....... . esz:2 .. zm:4 ....1. ..... ...0 . \
227
- &z2z_en n=4 zdn=%zd_ax4
229
+ &zzz_en n=4 zd=%zd_ax4 zn=%zd_ax4
230
231
SMAX_n1 1100000 1 .. 10 .... 1010.0 00000 .... 0 @z2z_2x1
232
SMAX_n1 1100000 1 .. 10 .... 1010.0 00000 .... 0 @z2z_4x1
@@ -990,8 +992,6 @@ UZP_2 11000001 esz:2 1 zm:5 110100 zn:5 .... 1 \
992
UZP_2 11000001 00 1 zm:5 110101 zn:5 .... 1 \
993
&zzz_e zd=%zd_ax2 esz=4
994
993
-&zzz_en zd zn zm esz n
994
-
995
FCLAMP 11000001 esz:2 1 zm:5 110000 zn:5 .... 0 \
996
&zzz_en zd=%zd_ax2 n=2
997
FCLAMP 11000001 esz:2 1 zm:5 110010 zn:5 ...0 0 \
target/arm/tcg/translate-sme.c
+35
-36
@@ -659,31 +659,30 @@ TRANS_FEAT(BMOPA, aa64_sme2, do_outprod, a, MO_32, gen_helper_sme2_bmopa_s)
659
TRANS_FEAT(SMOPA2_s, aa64_sme2, do_outprod, a, MO_32, gen_helper_sme2_smopa2_s)
660
TRANS_FEAT(UMOPA2_s, aa64_sme2, do_outprod, a, MO_32, gen_helper_sme2_umopa2_s)
661
662
-static bool do_z2z_n1(DisasContext *s, arg_z2z_en *a, GVecGen3Fn *fn)
662
+static bool do_zzz_n1(DisasContext *s, arg_zzz_en *a, GVecGen3Fn *fn)
663
{
664
- int esz, dn, vsz, mofs, n;
665
- bool overlap = false;
664
+ int esz = a->esz, vsz, mofs;
665
+ int overlap = -1;
666
667
if (!sme_sm_enabled_check(s)) {
668
return true;
669
}
670
671
- esz = a->esz;
672
- n = a->n;
673
- dn = a->zdn;
671
mofs = vec_full_reg_offset(s, a->zm);
672
vsz = streaming_vec_reg_size(s);
673
677
- for (int i = 0; i < n; i++) {
678
- int dofs = vec_full_reg_offset(s, dn + i);
674
+ for (int i = 0, n = a->n; i < n; i++) {
675
+ int dofs = vec_full_reg_offset(s, a->zd + i);
676
+ int nofs = vec_full_reg_offset(s, a->zn + i);
677
if (dofs == mofs) {
680
- overlap = true;
678
+ overlap = i;
679
} else {
682
- fn(esz, dofs, dofs, mofs, vsz, vsz);
680
+ fn(esz, dofs, nofs, mofs, vsz, vsz);
681
}
682
}
685
- if (overlap) {
686
- fn(esz, mofs, mofs, mofs, vsz, vsz);
683
+ if (overlap >= 0) {
684
+ int nofs = vec_full_reg_offset(s, a->zn + overlap);
685
+ fn(esz, mofs, nofs, mofs, vsz, vsz);
686
}
687
return true;
688
}
@@ -710,14 +709,14 @@ static void gen_sme2_urshl(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs,
709
tcg_gen_gvec_3_ool(rd_ofs, rn_ofs, rm_ofs, opr_sz, max_sz, 0, fns[vece]);
710
}
711
713
-TRANS_FEAT(ADD_n1, aa64_sme2, do_z2z_n1, a, tcg_gen_gvec_add)
714
-TRANS_FEAT(SMAX_n1, aa64_sme2, do_z2z_n1, a, tcg_gen_gvec_smax)
715
-TRANS_FEAT(SMIN_n1, aa64_sme2, do_z2z_n1, a, tcg_gen_gvec_smin)
716
-TRANS_FEAT(UMAX_n1, aa64_sme2, do_z2z_n1, a, tcg_gen_gvec_umax)
717
-TRANS_FEAT(UMIN_n1, aa64_sme2, do_z2z_n1, a, tcg_gen_gvec_umin)
718
-TRANS_FEAT(SRSHL_n1, aa64_sme2, do_z2z_n1, a, gen_sme2_srshl)
719
-TRANS_FEAT(URSHL_n1, aa64_sme2, do_z2z_n1, a, gen_sme2_urshl)
720
-TRANS_FEAT(SQDMULH_n1, aa64_sme2, do_z2z_n1, a, gen_gvec_sve2_sqdmulh)
712
+TRANS_FEAT(ADD_n1, aa64_sme2, do_zzz_n1, a, tcg_gen_gvec_add)
713
+TRANS_FEAT(SMAX_n1, aa64_sme2, do_zzz_n1, a, tcg_gen_gvec_smax)
714
+TRANS_FEAT(SMIN_n1, aa64_sme2, do_zzz_n1, a, tcg_gen_gvec_smin)
715
+TRANS_FEAT(UMAX_n1, aa64_sme2, do_zzz_n1, a, tcg_gen_gvec_umax)
716
+TRANS_FEAT(UMIN_n1, aa64_sme2, do_zzz_n1, a, tcg_gen_gvec_umin)
717
+TRANS_FEAT(SRSHL_n1, aa64_sme2, do_zzz_n1, a, gen_sme2_srshl)
718
+TRANS_FEAT(URSHL_n1, aa64_sme2, do_zzz_n1, a, gen_sme2_urshl)
719
+TRANS_FEAT(SQDMULH_n1, aa64_sme2, do_zzz_n1, a, gen_gvec_sve2_sqdmulh)
720
721
static bool do_z2z_nn(DisasContext *s, arg_z2z_en *a, GVecGen3Fn *fn)
722
{
@@ -750,11 +749,11 @@ TRANS_FEAT(SRSHL_nn, aa64_sme2, do_z2z_nn, a, gen_sme2_srshl)
749
TRANS_FEAT(URSHL_nn, aa64_sme2, do_z2z_nn, a, gen_sme2_urshl)
750
TRANS_FEAT(SQDMULH_nn, aa64_sme2, do_z2z_nn, a, gen_gvec_sve2_sqdmulh)
751
753
-static bool do_z2z_n1_fpst(DisasContext *s, arg_z2z_en *a,
752
+static bool do_zzz_n1_fpst(DisasContext *s, arg_zzz_en *a,
753
gen_helper_gvec_3_ptr * const fns[4])
754
{
756
- int esz = a->esz, n, dn, vsz, mofs;
757
- bool overlap = false;
755
+ int esz = a->esz, vsz, mofs;
756
+ int overlap = -1;
757
gen_helper_gvec_3_ptr *fn = fns[esz];
758
TCGv_ptr fpst;
759
@@ -770,21 +769,21 @@ static bool do_z2z_n1_fpst(DisasContext *s, arg_z2z_en *a,
769
}
770
771
fpst = fpstatus_ptr(esz == MO_16 ? FPST_A64_F16 : FPST_A64);
773
- n = a->n;
774
- dn = a->zdn;
772
mofs = vec_full_reg_offset(s, a->zm);
773
vsz = streaming_vec_reg_size(s);
774
778
- for (int i = 0; i < n; i++) {
779
- int dofs = vec_full_reg_offset(s, dn + i);
775
+ for (int i = 0, n = a->n; i < n; i++) {
776
+ int dofs = vec_full_reg_offset(s, a->zd + i);
777
+ int nofs = vec_full_reg_offset(s, a->zn + i);
778
if (dofs == mofs) {
781
- overlap = true;
779
+ overlap = i;
780
} else {
783
- tcg_gen_gvec_3_ptr(dofs, dofs, mofs, fpst, vsz, vsz, 0, fn);
781
+ tcg_gen_gvec_3_ptr(dofs, nofs, mofs, fpst, vsz, vsz, 0, fn);
782
}
783
}
786
- if (overlap) {
787
- tcg_gen_gvec_3_ptr(mofs, mofs, mofs, fpst, vsz, vsz, 0, fn);
784
+ if (overlap >= 0) {
785
+ int nofs = vec_full_reg_offset(s, a->zn + overlap);
786
+ tcg_gen_gvec_3_ptr(mofs, nofs, mofs, fpst, vsz, vsz, 0, fn);
787
}
788
return true;
789
}
@@ -831,7 +830,7 @@ static gen_helper_gvec_3_ptr * const f_vector_fmax[2][4] = {
830
gen_helper_gvec_ah_fmax_s,
831
gen_helper_gvec_ah_fmax_d },
832
};
834
-TRANS_FEAT(FMAX_n1, aa64_sme2, do_z2z_n1_fpst, a, f_vector_fmax[s->fpcr_ah])
833
+TRANS_FEAT(FMAX_n1, aa64_sme2, do_zzz_n1_fpst, a, f_vector_fmax[s->fpcr_ah])
834
TRANS_FEAT(FMAX_nn, aa64_sme2, do_z2z_nn_fpst, a, f_vector_fmax[s->fpcr_ah])
835
836
static gen_helper_gvec_3_ptr * const f_vector_fmin[2][4] = {
@@ -844,7 +843,7 @@ static gen_helper_gvec_3_ptr * const f_vector_fmin[2][4] = {
843
gen_helper_gvec_ah_fmin_s,
844
gen_helper_gvec_ah_fmin_d },
845
};
847
-TRANS_FEAT(FMIN_n1, aa64_sme2, do_z2z_n1_fpst, a, f_vector_fmin[s->fpcr_ah])
846
+TRANS_FEAT(FMIN_n1, aa64_sme2, do_zzz_n1_fpst, a, f_vector_fmin[s->fpcr_ah])
847
TRANS_FEAT(FMIN_nn, aa64_sme2, do_z2z_nn_fpst, a, f_vector_fmin[s->fpcr_ah])
848
849
static gen_helper_gvec_3_ptr * const f_vector_fmaxnm[4] = {
@@ -853,7 +852,7 @@ static gen_helper_gvec_3_ptr * const f_vector_fmaxnm[4] = {
852
gen_helper_gvec_fmaxnum_s,
853
gen_helper_gvec_fmaxnum_d,
854
};
856
-TRANS_FEAT(FMAXNM_n1, aa64_sme2, do_z2z_n1_fpst, a, f_vector_fmaxnm)
855
+TRANS_FEAT(FMAXNM_n1, aa64_sme2, do_zzz_n1_fpst, a, f_vector_fmaxnm)
856
TRANS_FEAT(FMAXNM_nn, aa64_sme2, do_z2z_nn_fpst, a, f_vector_fmaxnm)
857
858
static gen_helper_gvec_3_ptr * const f_vector_fminnm[4] = {
@@ -862,7 +861,7 @@ static gen_helper_gvec_3_ptr * const f_vector_fminnm[4] = {
861
gen_helper_gvec_fminnum_s,
862
gen_helper_gvec_fminnum_d,
863
};
865
-TRANS_FEAT(FMINNM_n1, aa64_sme2, do_z2z_n1_fpst, a, f_vector_fminnm)
864
+TRANS_FEAT(FMINNM_n1, aa64_sme2, do_zzz_n1_fpst, a, f_vector_fminnm)
865
TRANS_FEAT(FMINNM_nn, aa64_sme2, do_z2z_nn_fpst, a, f_vector_fminnm)
866
867
static gen_helper_gvec_3_ptr * const f_vector_famax[4] = {
@@ -887,7 +886,7 @@ static gen_helper_gvec_3_ptr * const f_vector_fscale[4] = {
886
gen_helper_gvec_fscale_s,
887
gen_helper_gvec_fscale_d,
888
};
890
-TRANS_FEAT(FSCALE_n1, aa64_sme2_f8cvt, do_z2z_n1_fpst, a, f_vector_fscale)
889
+TRANS_FEAT(FSCALE_n1, aa64_sme2_f8cvt, do_zzz_n1_fpst, a, f_vector_fscale)
890
TRANS_FEAT(FSCALE_nn, aa64_sme2_f8cvt, do_z2z_nn_fpst, a, f_vector_fscale)
891
892
/* Add/Sub vector Z[m] to each Z[n*N] with result in ZA[d*N]. */