2 * ARM translation: AArch32 Neon instructions
4 * Copyright (c) 2003 Fabrice Bellard
5 * Copyright (c) 2005-2007 CodeSourcery
6 * Copyright (c) 2007 OpenedHand, Ltd.
7 * Copyright (c) 2020 Linaro, Ltd.
9 * This library is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
14 * This library is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
23 #include "qemu/osdep.h"
24 #include "tcg/tcg-op.h"
25 #include "tcg/tcg-op-gvec.h"
26 #include "exec/exec-all.h"
27 #include "exec/gen-icount.h"
28 #include "translate.h"
29 #include "translate-a32.h"
31 static inline int plus1(DisasContext
*s
, int x
)
36 static inline int rsub_64(DisasContext
*s
, int x
)
41 static inline int rsub_32(DisasContext
*s
, int x
)
45 static inline int rsub_16(DisasContext
*s
, int x
)
49 static inline int rsub_8(DisasContext
*s
, int x
)
54 static inline int neon_3same_fp_size(DisasContext
*s
, int x
)
56 /* Convert 0==fp32, 1==fp16 into a MO_* value */
60 /* Include the generated Neon decoder */
61 #include "decode-neon-dp.c.inc"
62 #include "decode-neon-ls.c.inc"
63 #include "decode-neon-shared.c.inc"
65 static TCGv_ptr
vfp_reg_ptr(bool dp
, int reg
)
67 TCGv_ptr ret
= tcg_temp_new_ptr();
68 tcg_gen_addi_ptr(ret
, cpu_env
, vfp_reg_offset(dp
, reg
));
72 static void neon_load_element(TCGv_i32 var
, int reg
, int ele
, MemOp mop
)
74 long offset
= neon_element_offset(reg
, ele
, mop
& MO_SIZE
);
78 tcg_gen_ld8u_i32(var
, cpu_env
, offset
);
81 tcg_gen_ld16u_i32(var
, cpu_env
, offset
);
84 tcg_gen_ld_i32(var
, cpu_env
, offset
);
87 g_assert_not_reached();
91 static void neon_load_element64(TCGv_i64 var
, int reg
, int ele
, MemOp mop
)
93 long offset
= neon_element_offset(reg
, ele
, mop
& MO_SIZE
);
97 tcg_gen_ld8u_i64(var
, cpu_env
, offset
);
100 tcg_gen_ld16u_i64(var
, cpu_env
, offset
);
103 tcg_gen_ld32u_i64(var
, cpu_env
, offset
);
106 tcg_gen_ld_i64(var
, cpu_env
, offset
);
109 g_assert_not_reached();
113 static void neon_store_element(int reg
, int ele
, MemOp size
, TCGv_i32 var
)
115 long offset
= neon_element_offset(reg
, ele
, size
);
119 tcg_gen_st8_i32(var
, cpu_env
, offset
);
122 tcg_gen_st16_i32(var
, cpu_env
, offset
);
125 tcg_gen_st_i32(var
, cpu_env
, offset
);
128 g_assert_not_reached();
132 static void neon_store_element64(int reg
, int ele
, MemOp size
, TCGv_i64 var
)
134 long offset
= neon_element_offset(reg
, ele
, size
);
138 tcg_gen_st8_i64(var
, cpu_env
, offset
);
141 tcg_gen_st16_i64(var
, cpu_env
, offset
);
144 tcg_gen_st32_i64(var
, cpu_env
, offset
);
147 tcg_gen_st_i64(var
, cpu_env
, offset
);
150 g_assert_not_reached();
154 static bool trans_VCMLA(DisasContext
*s
, arg_VCMLA
*a
)
158 gen_helper_gvec_3_ptr
*fn_gvec_ptr
;
160 if (!dc_isar_feature(aa32_vcma
, s
)
161 || (a
->size
== MO_16
&& !dc_isar_feature(aa32_fp16_arith
, s
))) {
165 /* UNDEF accesses to D16-D31 if they don't exist. */
166 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
167 ((a
->vd
| a
->vn
| a
->vm
) & 0x10)) {
171 if ((a
->vn
| a
->vm
| a
->vd
) & a
->q
) {
175 if (!vfp_access_check(s
)) {
179 opr_sz
= (1 + a
->q
) * 8;
180 fpst
= fpstatus_ptr(a
->size
== MO_16
? FPST_STD_F16
: FPST_STD
);
181 fn_gvec_ptr
= (a
->size
== MO_16
) ?
182 gen_helper_gvec_fcmlah
: gen_helper_gvec_fcmlas
;
183 tcg_gen_gvec_3_ptr(vfp_reg_offset(1, a
->vd
),
184 vfp_reg_offset(1, a
->vn
),
185 vfp_reg_offset(1, a
->vm
),
186 fpst
, opr_sz
, opr_sz
, a
->rot
,
188 tcg_temp_free_ptr(fpst
);
192 static bool trans_VCADD(DisasContext
*s
, arg_VCADD
*a
)
196 gen_helper_gvec_3_ptr
*fn_gvec_ptr
;
198 if (!dc_isar_feature(aa32_vcma
, s
)
199 || (a
->size
== MO_16
&& !dc_isar_feature(aa32_fp16_arith
, s
))) {
203 /* UNDEF accesses to D16-D31 if they don't exist. */
204 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
205 ((a
->vd
| a
->vn
| a
->vm
) & 0x10)) {
209 if ((a
->vn
| a
->vm
| a
->vd
) & a
->q
) {
213 if (!vfp_access_check(s
)) {
217 opr_sz
= (1 + a
->q
) * 8;
218 fpst
= fpstatus_ptr(a
->size
== MO_16
? FPST_STD_F16
: FPST_STD
);
219 fn_gvec_ptr
= (a
->size
== MO_16
) ?
220 gen_helper_gvec_fcaddh
: gen_helper_gvec_fcadds
;
221 tcg_gen_gvec_3_ptr(vfp_reg_offset(1, a
->vd
),
222 vfp_reg_offset(1, a
->vn
),
223 vfp_reg_offset(1, a
->vm
),
224 fpst
, opr_sz
, opr_sz
, a
->rot
,
226 tcg_temp_free_ptr(fpst
);
230 static bool trans_VDOT(DisasContext
*s
, arg_VDOT
*a
)
233 gen_helper_gvec_3
*fn_gvec
;
235 if (!dc_isar_feature(aa32_dp
, s
)) {
239 /* UNDEF accesses to D16-D31 if they don't exist. */
240 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
241 ((a
->vd
| a
->vn
| a
->vm
) & 0x10)) {
245 if ((a
->vn
| a
->vm
| a
->vd
) & a
->q
) {
249 if (!vfp_access_check(s
)) {
253 opr_sz
= (1 + a
->q
) * 8;
254 fn_gvec
= a
->u
? gen_helper_gvec_udot_b
: gen_helper_gvec_sdot_b
;
255 tcg_gen_gvec_3_ool(vfp_reg_offset(1, a
->vd
),
256 vfp_reg_offset(1, a
->vn
),
257 vfp_reg_offset(1, a
->vm
),
258 opr_sz
, opr_sz
, 0, fn_gvec
);
262 static bool trans_VFML(DisasContext
*s
, arg_VFML
*a
)
266 if (!dc_isar_feature(aa32_fhm
, s
)) {
270 /* UNDEF accesses to D16-D31 if they don't exist. */
271 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
280 if (!vfp_access_check(s
)) {
284 opr_sz
= (1 + a
->q
) * 8;
285 tcg_gen_gvec_3_ptr(vfp_reg_offset(1, a
->vd
),
286 vfp_reg_offset(a
->q
, a
->vn
),
287 vfp_reg_offset(a
->q
, a
->vm
),
288 cpu_env
, opr_sz
, opr_sz
, a
->s
, /* is_2 == 0 */
289 gen_helper_gvec_fmlal_a32
);
293 static bool trans_VCMLA_scalar(DisasContext
*s
, arg_VCMLA_scalar
*a
)
295 gen_helper_gvec_3_ptr
*fn_gvec_ptr
;
299 if (!dc_isar_feature(aa32_vcma
, s
)) {
302 if (a
->size
== MO_16
&& !dc_isar_feature(aa32_fp16_arith
, s
)) {
306 /* UNDEF accesses to D16-D31 if they don't exist. */
307 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
308 ((a
->vd
| a
->vn
| a
->vm
) & 0x10)) {
312 if ((a
->vd
| a
->vn
) & a
->q
) {
316 if (!vfp_access_check(s
)) {
320 fn_gvec_ptr
= (a
->size
== MO_16
) ?
321 gen_helper_gvec_fcmlah_idx
: gen_helper_gvec_fcmlas_idx
;
322 opr_sz
= (1 + a
->q
) * 8;
323 fpst
= fpstatus_ptr(a
->size
== MO_16
? FPST_STD_F16
: FPST_STD
);
324 tcg_gen_gvec_3_ptr(vfp_reg_offset(1, a
->vd
),
325 vfp_reg_offset(1, a
->vn
),
326 vfp_reg_offset(1, a
->vm
),
327 fpst
, opr_sz
, opr_sz
,
328 (a
->index
<< 2) | a
->rot
, fn_gvec_ptr
);
329 tcg_temp_free_ptr(fpst
);
333 static bool trans_VDOT_scalar(DisasContext
*s
, arg_VDOT_scalar
*a
)
335 gen_helper_gvec_3
*fn_gvec
;
339 if (!dc_isar_feature(aa32_dp
, s
)) {
343 /* UNDEF accesses to D16-D31 if they don't exist. */
344 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
345 ((a
->vd
| a
->vn
) & 0x10)) {
349 if ((a
->vd
| a
->vn
) & a
->q
) {
353 if (!vfp_access_check(s
)) {
357 fn_gvec
= a
->u
? gen_helper_gvec_udot_idx_b
: gen_helper_gvec_sdot_idx_b
;
358 opr_sz
= (1 + a
->q
) * 8;
359 fpst
= fpstatus_ptr(FPST_STD
);
360 tcg_gen_gvec_3_ool(vfp_reg_offset(1, a
->vd
),
361 vfp_reg_offset(1, a
->vn
),
362 vfp_reg_offset(1, a
->rm
),
363 opr_sz
, opr_sz
, a
->index
, fn_gvec
);
364 tcg_temp_free_ptr(fpst
);
368 static bool trans_VFML_scalar(DisasContext
*s
, arg_VFML_scalar
*a
)
372 if (!dc_isar_feature(aa32_fhm
, s
)) {
376 /* UNDEF accesses to D16-D31 if they don't exist. */
377 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
378 ((a
->vd
& 0x10) || (a
->q
&& (a
->vn
& 0x10)))) {
386 if (!vfp_access_check(s
)) {
390 opr_sz
= (1 + a
->q
) * 8;
391 tcg_gen_gvec_3_ptr(vfp_reg_offset(1, a
->vd
),
392 vfp_reg_offset(a
->q
, a
->vn
),
393 vfp_reg_offset(a
->q
, a
->rm
),
394 cpu_env
, opr_sz
, opr_sz
,
395 (a
->index
<< 2) | a
->s
, /* is_2 == 0 */
396 gen_helper_gvec_fmlal_idx_a32
);
404 } const neon_ls_element_type
[11] = {
418 static void gen_neon_ldst_base_update(DisasContext
*s
, int rm
, int rn
,
424 base
= load_reg(s
, rn
);
426 tcg_gen_addi_i32(base
, base
, stride
);
429 index
= load_reg(s
, rm
);
430 tcg_gen_add_i32(base
, base
, index
);
431 tcg_temp_free_i32(index
);
433 store_reg(s
, rn
, base
);
437 static bool trans_VLDST_multiple(DisasContext
*s
, arg_VLDST_multiple
*a
)
439 /* Neon load/store multiple structures */
440 int nregs
, interleave
, spacing
, reg
, n
;
441 MemOp mop
, align
, endian
;
442 int mmu_idx
= get_mem_index(s
);
447 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
451 /* UNDEF accesses to D16-D31 if they don't exist */
452 if (!dc_isar_feature(aa32_simd_r32
, s
) && (a
->vd
& 0x10)) {
458 /* Catch UNDEF cases for bad values of align field */
459 switch (a
->itype
& 0xc) {
473 nregs
= neon_ls_element_type
[a
->itype
].nregs
;
474 interleave
= neon_ls_element_type
[a
->itype
].interleave
;
475 spacing
= neon_ls_element_type
[a
->itype
].spacing
;
476 if (size
== 3 && (interleave
| spacing
) != 1) {
480 if (!vfp_access_check(s
)) {
484 /* For our purposes, bytes are always little-endian. */
490 /* Enforce alignment requested by the instruction */
492 align
= pow2_align(a
->align
+ 2); /* 4 ** a->align */
494 align
= s
->align_mem
? MO_ALIGN
: 0;
498 * Consecutive little-endian elements from a single register
499 * can be promoted to a larger little-endian operation.
501 if (interleave
== 1 && endian
== MO_LE
) {
502 /* Retain any natural alignment. */
503 if (align
== MO_ALIGN
) {
504 align
= pow2_align(size
);
509 tmp64
= tcg_temp_new_i64();
510 addr
= tcg_temp_new_i32();
511 tmp
= tcg_const_i32(1 << size
);
512 load_reg_var(s
, addr
, a
->rn
);
514 mop
= endian
| size
| align
;
515 for (reg
= 0; reg
< nregs
; reg
++) {
516 for (n
= 0; n
< 8 >> size
; n
++) {
518 for (xs
= 0; xs
< interleave
; xs
++) {
519 int tt
= a
->vd
+ reg
+ spacing
* xs
;
522 gen_aa32_ld_internal_i64(s
, tmp64
, addr
, mmu_idx
, mop
);
523 neon_store_element64(tt
, n
, size
, tmp64
);
525 neon_load_element64(tmp64
, tt
, n
, size
);
526 gen_aa32_st_internal_i64(s
, tmp64
, addr
, mmu_idx
, mop
);
528 tcg_gen_add_i32(addr
, addr
, tmp
);
530 /* Subsequent memory operations inherit alignment */
535 tcg_temp_free_i32(addr
);
536 tcg_temp_free_i32(tmp
);
537 tcg_temp_free_i64(tmp64
);
539 gen_neon_ldst_base_update(s
, a
->rm
, a
->rn
, nregs
* interleave
* 8);
543 static bool trans_VLD_all_lanes(DisasContext
*s
, arg_VLD_all_lanes
*a
)
545 /* Neon load single structure to all lanes */
546 int reg
, stride
, vec_size
;
549 int nregs
= a
->n
+ 1;
553 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
557 /* UNDEF accesses to D16-D31 if they don't exist */
558 if (!dc_isar_feature(aa32_simd_r32
, s
) && (a
->vd
& 0x10)) {
564 if (nregs
!= 4 || a
->a
== 0) {
567 /* For VLD4 size == 3 a == 1 means 32 bits at 16 byte alignment */
579 align
= pow2_align(size
+ 1);
584 align
= pow2_align(size
+ 2);
587 g_assert_not_reached();
591 if (!vfp_access_check(s
)) {
596 * VLD1 to all lanes: T bit indicates how many Dregs to write.
597 * VLD2/3/4 to all lanes: T bit indicates register stride.
599 stride
= a
->t
? 2 : 1;
600 vec_size
= nregs
== 1 ? stride
* 8 : 8;
602 tmp
= tcg_temp_new_i32();
603 addr
= tcg_temp_new_i32();
604 load_reg_var(s
, addr
, a
->rn
);
605 for (reg
= 0; reg
< nregs
; reg
++) {
606 gen_aa32_ld_i32(s
, tmp
, addr
, get_mem_index(s
), mop
);
607 if ((vd
& 1) && vec_size
== 16) {
609 * We cannot write 16 bytes at once because the
610 * destination is unaligned.
612 tcg_gen_gvec_dup_i32(size
, neon_full_reg_offset(vd
),
614 tcg_gen_gvec_mov(0, neon_full_reg_offset(vd
+ 1),
615 neon_full_reg_offset(vd
), 8, 8);
617 tcg_gen_gvec_dup_i32(size
, neon_full_reg_offset(vd
),
618 vec_size
, vec_size
, tmp
);
620 tcg_gen_addi_i32(addr
, addr
, 1 << size
);
623 /* Subsequent memory operations inherit alignment */
626 tcg_temp_free_i32(tmp
);
627 tcg_temp_free_i32(addr
);
629 gen_neon_ldst_base_update(s
, a
->rm
, a
->rn
, (1 << size
) * nregs
);
634 static bool trans_VLDST_single(DisasContext
*s
, arg_VLDST_single
*a
)
636 /* Neon load/store single structure to one lane */
638 int nregs
= a
->n
+ 1;
643 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
647 /* UNDEF accesses to D16-D31 if they don't exist */
648 if (!dc_isar_feature(aa32_simd_r32
, s
) && (a
->vd
& 0x10)) {
652 /* Catch the UNDEF cases. This is unavoidably a bit messy. */
655 if (((a
->align
& (1 << a
->size
)) != 0) ||
656 (a
->size
== 2 && (a
->align
== 1 || a
->align
== 2))) {
661 if ((a
->align
& 1) != 0) {
666 if (a
->size
== 2 && (a
->align
& 2) != 0) {
671 if (a
->size
== 2 && a
->align
== 3) {
678 if ((vd
+ a
->stride
* (nregs
- 1)) > 31) {
680 * Attempts to write off the end of the register file are
681 * UNPREDICTABLE; we choose to UNDEF because otherwise we would
682 * access off the end of the array that holds the register data.
687 if (!vfp_access_check(s
)) {
691 /* Pick up SCTLR settings */
692 mop
= finalize_memop(s
, a
->size
);
699 /* For VLD1, use natural alignment. */
703 /* For VLD2, use double alignment. */
704 align_op
= pow2_align(a
->size
+ 1);
707 if (a
->size
== MO_32
) {
709 * For VLD4.32, align = 1 is double alignment, align = 2 is
710 * quad alignment; align = 3 is rejected above.
712 align_op
= pow2_align(a
->size
+ a
->align
);
714 /* For VLD4.8 and VLD.16, we want quad alignment. */
715 align_op
= pow2_align(a
->size
+ 2);
719 /* For VLD3, the alignment field is zero and rejected above. */
720 g_assert_not_reached();
723 mop
= (mop
& ~MO_AMASK
) | align_op
;
726 tmp
= tcg_temp_new_i32();
727 addr
= tcg_temp_new_i32();
728 load_reg_var(s
, addr
, a
->rn
);
730 for (reg
= 0; reg
< nregs
; reg
++) {
732 gen_aa32_ld_internal_i32(s
, tmp
, addr
, get_mem_index(s
), mop
);
733 neon_store_element(vd
, a
->reg_idx
, a
->size
, tmp
);
735 neon_load_element(tmp
, vd
, a
->reg_idx
, a
->size
);
736 gen_aa32_st_internal_i32(s
, tmp
, addr
, get_mem_index(s
), mop
);
739 tcg_gen_addi_i32(addr
, addr
, 1 << a
->size
);
741 /* Subsequent memory operations inherit alignment */
744 tcg_temp_free_i32(addr
);
745 tcg_temp_free_i32(tmp
);
747 gen_neon_ldst_base_update(s
, a
->rm
, a
->rn
, (1 << a
->size
) * nregs
);
752 static bool do_3same(DisasContext
*s
, arg_3same
*a
, GVecGen3Fn fn
)
754 int vec_size
= a
->q
? 16 : 8;
755 int rd_ofs
= neon_full_reg_offset(a
->vd
);
756 int rn_ofs
= neon_full_reg_offset(a
->vn
);
757 int rm_ofs
= neon_full_reg_offset(a
->vm
);
759 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
763 /* UNDEF accesses to D16-D31 if they don't exist. */
764 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
765 ((a
->vd
| a
->vn
| a
->vm
) & 0x10)) {
769 if ((a
->vn
| a
->vm
| a
->vd
) & a
->q
) {
773 if (!vfp_access_check(s
)) {
777 fn(a
->size
, rd_ofs
, rn_ofs
, rm_ofs
, vec_size
, vec_size
);
781 #define DO_3SAME(INSN, FUNC) \
782 static bool trans_##INSN##_3s(DisasContext *s, arg_3same *a) \
784 return do_3same(s, a, FUNC); \
787 DO_3SAME(VADD
, tcg_gen_gvec_add
)
788 DO_3SAME(VSUB
, tcg_gen_gvec_sub
)
789 DO_3SAME(VAND
, tcg_gen_gvec_and
)
790 DO_3SAME(VBIC
, tcg_gen_gvec_andc
)
791 DO_3SAME(VORR
, tcg_gen_gvec_or
)
792 DO_3SAME(VORN
, tcg_gen_gvec_orc
)
793 DO_3SAME(VEOR
, tcg_gen_gvec_xor
)
794 DO_3SAME(VSHL_S
, gen_gvec_sshl
)
795 DO_3SAME(VSHL_U
, gen_gvec_ushl
)
796 DO_3SAME(VQADD_S
, gen_gvec_sqadd_qc
)
797 DO_3SAME(VQADD_U
, gen_gvec_uqadd_qc
)
798 DO_3SAME(VQSUB_S
, gen_gvec_sqsub_qc
)
799 DO_3SAME(VQSUB_U
, gen_gvec_uqsub_qc
)
801 /* These insns are all gvec_bitsel but with the inputs in various orders. */
802 #define DO_3SAME_BITSEL(INSN, O1, O2, O3) \
803 static void gen_##INSN##_3s(unsigned vece, uint32_t rd_ofs, \
804 uint32_t rn_ofs, uint32_t rm_ofs, \
805 uint32_t oprsz, uint32_t maxsz) \
807 tcg_gen_gvec_bitsel(vece, rd_ofs, O1, O2, O3, oprsz, maxsz); \
809 DO_3SAME(INSN, gen_##INSN##_3s)
811 DO_3SAME_BITSEL(VBSL
, rd_ofs
, rn_ofs
, rm_ofs
)
812 DO_3SAME_BITSEL(VBIT
, rm_ofs
, rn_ofs
, rd_ofs
)
813 DO_3SAME_BITSEL(VBIF
, rm_ofs
, rd_ofs
, rn_ofs
)
815 #define DO_3SAME_NO_SZ_3(INSN, FUNC) \
816 static bool trans_##INSN##_3s(DisasContext *s, arg_3same *a) \
818 if (a->size == 3) { \
821 return do_3same(s, a, FUNC); \
824 DO_3SAME_NO_SZ_3(VMAX_S
, tcg_gen_gvec_smax
)
825 DO_3SAME_NO_SZ_3(VMAX_U
, tcg_gen_gvec_umax
)
826 DO_3SAME_NO_SZ_3(VMIN_S
, tcg_gen_gvec_smin
)
827 DO_3SAME_NO_SZ_3(VMIN_U
, tcg_gen_gvec_umin
)
828 DO_3SAME_NO_SZ_3(VMUL
, tcg_gen_gvec_mul
)
829 DO_3SAME_NO_SZ_3(VMLA
, gen_gvec_mla
)
830 DO_3SAME_NO_SZ_3(VMLS
, gen_gvec_mls
)
831 DO_3SAME_NO_SZ_3(VTST
, gen_gvec_cmtst
)
832 DO_3SAME_NO_SZ_3(VABD_S
, gen_gvec_sabd
)
833 DO_3SAME_NO_SZ_3(VABA_S
, gen_gvec_saba
)
834 DO_3SAME_NO_SZ_3(VABD_U
, gen_gvec_uabd
)
835 DO_3SAME_NO_SZ_3(VABA_U
, gen_gvec_uaba
)
837 #define DO_3SAME_CMP(INSN, COND) \
838 static void gen_##INSN##_3s(unsigned vece, uint32_t rd_ofs, \
839 uint32_t rn_ofs, uint32_t rm_ofs, \
840 uint32_t oprsz, uint32_t maxsz) \
842 tcg_gen_gvec_cmp(COND, vece, rd_ofs, rn_ofs, rm_ofs, oprsz, maxsz); \
844 DO_3SAME_NO_SZ_3(INSN, gen_##INSN##_3s)
846 DO_3SAME_CMP(VCGT_S
, TCG_COND_GT
)
847 DO_3SAME_CMP(VCGT_U
, TCG_COND_GTU
)
848 DO_3SAME_CMP(VCGE_S
, TCG_COND_GE
)
849 DO_3SAME_CMP(VCGE_U
, TCG_COND_GEU
)
850 DO_3SAME_CMP(VCEQ
, TCG_COND_EQ
)
852 #define WRAP_OOL_FN(WRAPNAME, FUNC) \
853 static void WRAPNAME(unsigned vece, uint32_t rd_ofs, uint32_t rn_ofs, \
854 uint32_t rm_ofs, uint32_t oprsz, uint32_t maxsz) \
856 tcg_gen_gvec_3_ool(rd_ofs, rn_ofs, rm_ofs, oprsz, maxsz, 0, FUNC); \
859 WRAP_OOL_FN(gen_VMUL_p_3s
, gen_helper_gvec_pmul_b
)
861 static bool trans_VMUL_p_3s(DisasContext
*s
, arg_3same
*a
)
866 return do_3same(s
, a
, gen_VMUL_p_3s
);
869 #define DO_VQRDMLAH(INSN, FUNC) \
870 static bool trans_##INSN##_3s(DisasContext *s, arg_3same *a) \
872 if (!dc_isar_feature(aa32_rdm, s)) { \
875 if (a->size != 1 && a->size != 2) { \
878 return do_3same(s, a, FUNC); \
881 DO_VQRDMLAH(VQRDMLAH
, gen_gvec_sqrdmlah_qc
)
882 DO_VQRDMLAH(VQRDMLSH
, gen_gvec_sqrdmlsh_qc
)
884 #define DO_SHA1(NAME, FUNC) \
885 WRAP_OOL_FN(gen_##NAME##_3s, FUNC) \
886 static bool trans_##NAME##_3s(DisasContext *s, arg_3same *a) \
888 if (!dc_isar_feature(aa32_sha1, s)) { \
891 return do_3same(s, a, gen_##NAME##_3s); \
894 DO_SHA1(SHA1C
, gen_helper_crypto_sha1c
)
895 DO_SHA1(SHA1P
, gen_helper_crypto_sha1p
)
896 DO_SHA1(SHA1M
, gen_helper_crypto_sha1m
)
897 DO_SHA1(SHA1SU0
, gen_helper_crypto_sha1su0
)
899 #define DO_SHA2(NAME, FUNC) \
900 WRAP_OOL_FN(gen_##NAME##_3s, FUNC) \
901 static bool trans_##NAME##_3s(DisasContext *s, arg_3same *a) \
903 if (!dc_isar_feature(aa32_sha2, s)) { \
906 return do_3same(s, a, gen_##NAME##_3s); \
909 DO_SHA2(SHA256H
, gen_helper_crypto_sha256h
)
910 DO_SHA2(SHA256H2
, gen_helper_crypto_sha256h2
)
911 DO_SHA2(SHA256SU1
, gen_helper_crypto_sha256su1
)
913 #define DO_3SAME_64(INSN, FUNC) \
914 static void gen_##INSN##_3s(unsigned vece, uint32_t rd_ofs, \
915 uint32_t rn_ofs, uint32_t rm_ofs, \
916 uint32_t oprsz, uint32_t maxsz) \
918 static const GVecGen3 op = { .fni8 = FUNC }; \
919 tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, oprsz, maxsz, &op); \
921 DO_3SAME(INSN, gen_##INSN##_3s)
923 #define DO_3SAME_64_ENV(INSN, FUNC) \
924 static void gen_##INSN##_elt(TCGv_i64 d, TCGv_i64 n, TCGv_i64 m) \
926 FUNC(d, cpu_env, n, m); \
928 DO_3SAME_64(INSN, gen_##INSN##_elt)
930 DO_3SAME_64(VRSHL_S64
, gen_helper_neon_rshl_s64
)
931 DO_3SAME_64(VRSHL_U64
, gen_helper_neon_rshl_u64
)
932 DO_3SAME_64_ENV(VQSHL_S64
, gen_helper_neon_qshl_s64
)
933 DO_3SAME_64_ENV(VQSHL_U64
, gen_helper_neon_qshl_u64
)
934 DO_3SAME_64_ENV(VQRSHL_S64
, gen_helper_neon_qrshl_s64
)
935 DO_3SAME_64_ENV(VQRSHL_U64
, gen_helper_neon_qrshl_u64
)
937 #define DO_3SAME_32(INSN, FUNC) \
938 static void gen_##INSN##_3s(unsigned vece, uint32_t rd_ofs, \
939 uint32_t rn_ofs, uint32_t rm_ofs, \
940 uint32_t oprsz, uint32_t maxsz) \
942 static const GVecGen3 ops[4] = { \
943 { .fni4 = gen_helper_neon_##FUNC##8 }, \
944 { .fni4 = gen_helper_neon_##FUNC##16 }, \
945 { .fni4 = gen_helper_neon_##FUNC##32 }, \
948 tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, oprsz, maxsz, &ops[vece]); \
950 static bool trans_##INSN##_3s(DisasContext *s, arg_3same *a) \
955 return do_3same(s, a, gen_##INSN##_3s); \
959 * Some helper functions need to be passed the cpu_env. In order
960 * to use those with the gvec APIs like tcg_gen_gvec_3() we need
961 * to create wrapper functions whose prototype is a NeonGenTwoOpFn()
962 * and which call a NeonGenTwoOpEnvFn().
964 #define WRAP_ENV_FN(WRAPNAME, FUNC) \
965 static void WRAPNAME(TCGv_i32 d, TCGv_i32 n, TCGv_i32 m) \
967 FUNC(d, cpu_env, n, m); \
970 #define DO_3SAME_32_ENV(INSN, FUNC) \
971 WRAP_ENV_FN(gen_##INSN##_tramp8, gen_helper_neon_##FUNC##8); \
972 WRAP_ENV_FN(gen_##INSN##_tramp16, gen_helper_neon_##FUNC##16); \
973 WRAP_ENV_FN(gen_##INSN##_tramp32, gen_helper_neon_##FUNC##32); \
974 static void gen_##INSN##_3s(unsigned vece, uint32_t rd_ofs, \
975 uint32_t rn_ofs, uint32_t rm_ofs, \
976 uint32_t oprsz, uint32_t maxsz) \
978 static const GVecGen3 ops[4] = { \
979 { .fni4 = gen_##INSN##_tramp8 }, \
980 { .fni4 = gen_##INSN##_tramp16 }, \
981 { .fni4 = gen_##INSN##_tramp32 }, \
984 tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, oprsz, maxsz, &ops[vece]); \
986 static bool trans_##INSN##_3s(DisasContext *s, arg_3same *a) \
991 return do_3same(s, a, gen_##INSN##_3s); \
994 DO_3SAME_32(VHADD_S
, hadd_s
)
995 DO_3SAME_32(VHADD_U
, hadd_u
)
996 DO_3SAME_32(VHSUB_S
, hsub_s
)
997 DO_3SAME_32(VHSUB_U
, hsub_u
)
998 DO_3SAME_32(VRHADD_S
, rhadd_s
)
999 DO_3SAME_32(VRHADD_U
, rhadd_u
)
1000 DO_3SAME_32(VRSHL_S
, rshl_s
)
1001 DO_3SAME_32(VRSHL_U
, rshl_u
)
1003 DO_3SAME_32_ENV(VQSHL_S
, qshl_s
)
1004 DO_3SAME_32_ENV(VQSHL_U
, qshl_u
)
1005 DO_3SAME_32_ENV(VQRSHL_S
, qrshl_s
)
1006 DO_3SAME_32_ENV(VQRSHL_U
, qrshl_u
)
1008 static bool do_3same_pair(DisasContext
*s
, arg_3same
*a
, NeonGenTwoOpFn
*fn
)
1010 /* Operations handled pairwise 32 bits at a time */
1011 TCGv_i32 tmp
, tmp2
, tmp3
;
1013 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
1017 /* UNDEF accesses to D16-D31 if they don't exist. */
1018 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
1019 ((a
->vd
| a
->vn
| a
->vm
) & 0x10)) {
1027 if (!vfp_access_check(s
)) {
1031 assert(a
->q
== 0); /* enforced by decode patterns */
1034 * Note that we have to be careful not to clobber the source operands
1035 * in the "vm == vd" case by storing the result of the first pass too
1036 * early. Since Q is 0 there are always just two passes, so instead
1037 * of a complicated loop over each pass we just unroll.
1039 tmp
= tcg_temp_new_i32();
1040 tmp2
= tcg_temp_new_i32();
1041 tmp3
= tcg_temp_new_i32();
1043 read_neon_element32(tmp
, a
->vn
, 0, MO_32
);
1044 read_neon_element32(tmp2
, a
->vn
, 1, MO_32
);
1047 read_neon_element32(tmp3
, a
->vm
, 0, MO_32
);
1048 read_neon_element32(tmp2
, a
->vm
, 1, MO_32
);
1049 fn(tmp3
, tmp3
, tmp2
);
1051 write_neon_element32(tmp
, a
->vd
, 0, MO_32
);
1052 write_neon_element32(tmp3
, a
->vd
, 1, MO_32
);
1054 tcg_temp_free_i32(tmp
);
1055 tcg_temp_free_i32(tmp2
);
1056 tcg_temp_free_i32(tmp3
);
1060 #define DO_3SAME_PAIR(INSN, func) \
1061 static bool trans_##INSN##_3s(DisasContext *s, arg_3same *a) \
1063 static NeonGenTwoOpFn * const fns[] = { \
1064 gen_helper_neon_##func##8, \
1065 gen_helper_neon_##func##16, \
1066 gen_helper_neon_##func##32, \
1068 if (a->size > 2) { \
1071 return do_3same_pair(s, a, fns[a->size]); \
1074 /* 32-bit pairwise ops end up the same as the elementwise versions. */
1075 #define gen_helper_neon_pmax_s32 tcg_gen_smax_i32
1076 #define gen_helper_neon_pmax_u32 tcg_gen_umax_i32
1077 #define gen_helper_neon_pmin_s32 tcg_gen_smin_i32
1078 #define gen_helper_neon_pmin_u32 tcg_gen_umin_i32
1079 #define gen_helper_neon_padd_u32 tcg_gen_add_i32
1081 DO_3SAME_PAIR(VPMAX_S
, pmax_s
)
1082 DO_3SAME_PAIR(VPMIN_S
, pmin_s
)
1083 DO_3SAME_PAIR(VPMAX_U
, pmax_u
)
1084 DO_3SAME_PAIR(VPMIN_U
, pmin_u
)
1085 DO_3SAME_PAIR(VPADD
, padd_u
)
1087 #define DO_3SAME_VQDMULH(INSN, FUNC) \
1088 WRAP_ENV_FN(gen_##INSN##_tramp16, gen_helper_neon_##FUNC##_s16); \
1089 WRAP_ENV_FN(gen_##INSN##_tramp32, gen_helper_neon_##FUNC##_s32); \
1090 static void gen_##INSN##_3s(unsigned vece, uint32_t rd_ofs, \
1091 uint32_t rn_ofs, uint32_t rm_ofs, \
1092 uint32_t oprsz, uint32_t maxsz) \
1094 static const GVecGen3 ops[2] = { \
1095 { .fni4 = gen_##INSN##_tramp16 }, \
1096 { .fni4 = gen_##INSN##_tramp32 }, \
1098 tcg_gen_gvec_3(rd_ofs, rn_ofs, rm_ofs, oprsz, maxsz, &ops[vece - 1]); \
1100 static bool trans_##INSN##_3s(DisasContext *s, arg_3same *a) \
1102 if (a->size != 1 && a->size != 2) { \
1105 return do_3same(s, a, gen_##INSN##_3s); \
1108 DO_3SAME_VQDMULH(VQDMULH
, qdmulh
)
1109 DO_3SAME_VQDMULH(VQRDMULH
, qrdmulh
)
1111 #define WRAP_FP_GVEC(WRAPNAME, FPST, FUNC) \
1112 static void WRAPNAME(unsigned vece, uint32_t rd_ofs, \
1113 uint32_t rn_ofs, uint32_t rm_ofs, \
1114 uint32_t oprsz, uint32_t maxsz) \
1116 TCGv_ptr fpst = fpstatus_ptr(FPST); \
1117 tcg_gen_gvec_3_ptr(rd_ofs, rn_ofs, rm_ofs, fpst, \
1118 oprsz, maxsz, 0, FUNC); \
1119 tcg_temp_free_ptr(fpst); \
1122 #define DO_3S_FP_GVEC(INSN,SFUNC,HFUNC) \
1123 WRAP_FP_GVEC(gen_##INSN##_fp32_3s, FPST_STD, SFUNC) \
1124 WRAP_FP_GVEC(gen_##INSN##_fp16_3s, FPST_STD_F16, HFUNC) \
1125 static bool trans_##INSN##_fp_3s(DisasContext *s, arg_3same *a) \
1127 if (a->size == MO_16) { \
1128 if (!dc_isar_feature(aa32_fp16_arith, s)) { \
1131 return do_3same(s, a, gen_##INSN##_fp16_3s); \
1133 return do_3same(s, a, gen_##INSN##_fp32_3s); \
1137 DO_3S_FP_GVEC(VADD
, gen_helper_gvec_fadd_s
, gen_helper_gvec_fadd_h
)
1138 DO_3S_FP_GVEC(VSUB
, gen_helper_gvec_fsub_s
, gen_helper_gvec_fsub_h
)
1139 DO_3S_FP_GVEC(VABD
, gen_helper_gvec_fabd_s
, gen_helper_gvec_fabd_h
)
1140 DO_3S_FP_GVEC(VMUL
, gen_helper_gvec_fmul_s
, gen_helper_gvec_fmul_h
)
1141 DO_3S_FP_GVEC(VCEQ
, gen_helper_gvec_fceq_s
, gen_helper_gvec_fceq_h
)
1142 DO_3S_FP_GVEC(VCGE
, gen_helper_gvec_fcge_s
, gen_helper_gvec_fcge_h
)
1143 DO_3S_FP_GVEC(VCGT
, gen_helper_gvec_fcgt_s
, gen_helper_gvec_fcgt_h
)
1144 DO_3S_FP_GVEC(VACGE
, gen_helper_gvec_facge_s
, gen_helper_gvec_facge_h
)
1145 DO_3S_FP_GVEC(VACGT
, gen_helper_gvec_facgt_s
, gen_helper_gvec_facgt_h
)
1146 DO_3S_FP_GVEC(VMAX
, gen_helper_gvec_fmax_s
, gen_helper_gvec_fmax_h
)
1147 DO_3S_FP_GVEC(VMIN
, gen_helper_gvec_fmin_s
, gen_helper_gvec_fmin_h
)
1148 DO_3S_FP_GVEC(VMLA
, gen_helper_gvec_fmla_s
, gen_helper_gvec_fmla_h
)
1149 DO_3S_FP_GVEC(VMLS
, gen_helper_gvec_fmls_s
, gen_helper_gvec_fmls_h
)
1150 DO_3S_FP_GVEC(VFMA
, gen_helper_gvec_vfma_s
, gen_helper_gvec_vfma_h
)
1151 DO_3S_FP_GVEC(VFMS
, gen_helper_gvec_vfms_s
, gen_helper_gvec_vfms_h
)
1152 DO_3S_FP_GVEC(VRECPS
, gen_helper_gvec_recps_nf_s
, gen_helper_gvec_recps_nf_h
)
1153 DO_3S_FP_GVEC(VRSQRTS
, gen_helper_gvec_rsqrts_nf_s
, gen_helper_gvec_rsqrts_nf_h
)
1155 WRAP_FP_GVEC(gen_VMAXNM_fp32_3s
, FPST_STD
, gen_helper_gvec_fmaxnum_s
)
1156 WRAP_FP_GVEC(gen_VMAXNM_fp16_3s
, FPST_STD_F16
, gen_helper_gvec_fmaxnum_h
)
1157 WRAP_FP_GVEC(gen_VMINNM_fp32_3s
, FPST_STD
, gen_helper_gvec_fminnum_s
)
1158 WRAP_FP_GVEC(gen_VMINNM_fp16_3s
, FPST_STD_F16
, gen_helper_gvec_fminnum_h
)
1160 static bool trans_VMAXNM_fp_3s(DisasContext
*s
, arg_3same
*a
)
1162 if (!arm_dc_feature(s
, ARM_FEATURE_V8
)) {
1166 if (a
->size
== MO_16
) {
1167 if (!dc_isar_feature(aa32_fp16_arith
, s
)) {
1170 return do_3same(s
, a
, gen_VMAXNM_fp16_3s
);
1172 return do_3same(s
, a
, gen_VMAXNM_fp32_3s
);
1175 static bool trans_VMINNM_fp_3s(DisasContext
*s
, arg_3same
*a
)
1177 if (!arm_dc_feature(s
, ARM_FEATURE_V8
)) {
1181 if (a
->size
== MO_16
) {
1182 if (!dc_isar_feature(aa32_fp16_arith
, s
)) {
1185 return do_3same(s
, a
, gen_VMINNM_fp16_3s
);
1187 return do_3same(s
, a
, gen_VMINNM_fp32_3s
);
1190 static bool do_3same_fp_pair(DisasContext
*s
, arg_3same
*a
,
1191 gen_helper_gvec_3_ptr
*fn
)
1193 /* FP pairwise operations */
1196 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
1200 /* UNDEF accesses to D16-D31 if they don't exist. */
1201 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
1202 ((a
->vd
| a
->vn
| a
->vm
) & 0x10)) {
1206 if (!vfp_access_check(s
)) {
1210 assert(a
->q
== 0); /* enforced by decode patterns */
1213 fpstatus
= fpstatus_ptr(a
->size
== MO_16
? FPST_STD_F16
: FPST_STD
);
1214 tcg_gen_gvec_3_ptr(vfp_reg_offset(1, a
->vd
),
1215 vfp_reg_offset(1, a
->vn
),
1216 vfp_reg_offset(1, a
->vm
),
1217 fpstatus
, 8, 8, 0, fn
);
1218 tcg_temp_free_ptr(fpstatus
);
1224 * For all the functions using this macro, size == 1 means fp16,
1225 * which is an architecture extension we don't implement yet.
1227 #define DO_3S_FP_PAIR(INSN,FUNC) \
1228 static bool trans_##INSN##_fp_3s(DisasContext *s, arg_3same *a) \
1230 if (a->size == MO_16) { \
1231 if (!dc_isar_feature(aa32_fp16_arith, s)) { \
1234 return do_3same_fp_pair(s, a, FUNC##h); \
1236 return do_3same_fp_pair(s, a, FUNC##s); \
1239 DO_3S_FP_PAIR(VPADD
, gen_helper_neon_padd
)
1240 DO_3S_FP_PAIR(VPMAX
, gen_helper_neon_pmax
)
1241 DO_3S_FP_PAIR(VPMIN
, gen_helper_neon_pmin
)
1243 static bool do_vector_2sh(DisasContext
*s
, arg_2reg_shift
*a
, GVecGen2iFn
*fn
)
1245 /* Handle a 2-reg-shift insn which can be vectorized. */
1246 int vec_size
= a
->q
? 16 : 8;
1247 int rd_ofs
= neon_full_reg_offset(a
->vd
);
1248 int rm_ofs
= neon_full_reg_offset(a
->vm
);
1250 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
1254 /* UNDEF accesses to D16-D31 if they don't exist. */
1255 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
1256 ((a
->vd
| a
->vm
) & 0x10)) {
1260 if ((a
->vm
| a
->vd
) & a
->q
) {
1264 if (!vfp_access_check(s
)) {
1268 fn(a
->size
, rd_ofs
, rm_ofs
, a
->shift
, vec_size
, vec_size
);
1272 #define DO_2SH(INSN, FUNC) \
1273 static bool trans_##INSN##_2sh(DisasContext *s, arg_2reg_shift *a) \
1275 return do_vector_2sh(s, a, FUNC); \
1278 DO_2SH(VSHL, tcg_gen_gvec_shli)
1279 DO_2SH(VSLI
, gen_gvec_sli
)
1280 DO_2SH(VSRI
, gen_gvec_sri
)
1281 DO_2SH(VSRA_S
, gen_gvec_ssra
)
1282 DO_2SH(VSRA_U
, gen_gvec_usra
)
1283 DO_2SH(VRSHR_S
, gen_gvec_srshr
)
1284 DO_2SH(VRSHR_U
, gen_gvec_urshr
)
1285 DO_2SH(VRSRA_S
, gen_gvec_srsra
)
1286 DO_2SH(VRSRA_U
, gen_gvec_ursra
)
1288 static bool trans_VSHR_S_2sh(DisasContext
*s
, arg_2reg_shift
*a
)
1290 /* Signed shift out of range results in all-sign-bits */
1291 a
->shift
= MIN(a
->shift
, (8 << a
->size
) - 1);
1292 return do_vector_2sh(s
, a
, tcg_gen_gvec_sari
);
1295 static void gen_zero_rd_2sh(unsigned vece
, uint32_t rd_ofs
, uint32_t rm_ofs
,
1296 int64_t shift
, uint32_t oprsz
, uint32_t maxsz
)
1298 tcg_gen_gvec_dup_imm(vece
, rd_ofs
, oprsz
, maxsz
, 0);
1301 static bool trans_VSHR_U_2sh(DisasContext
*s
, arg_2reg_shift
*a
)
1303 /* Shift out of range is architecturally valid and results in zero. */
1304 if (a
->shift
>= (8 << a
->size
)) {
1305 return do_vector_2sh(s
, a
, gen_zero_rd_2sh
);
1307 return do_vector_2sh(s
, a
, tcg_gen_gvec_shri
);
1311 static bool do_2shift_env_64(DisasContext
*s
, arg_2reg_shift
*a
,
1312 NeonGenTwo64OpEnvFn
*fn
)
1315 * 2-reg-and-shift operations, size == 3 case, where the
1316 * function needs to be passed cpu_env.
1321 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
1325 /* UNDEF accesses to D16-D31 if they don't exist. */
1326 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
1327 ((a
->vd
| a
->vm
) & 0x10)) {
1331 if ((a
->vm
| a
->vd
) & a
->q
) {
1335 if (!vfp_access_check(s
)) {
1340 * To avoid excessive duplication of ops we implement shift
1341 * by immediate using the variable shift operations.
1343 constimm
= tcg_const_i64(dup_const(a
->size
, a
->shift
));
1345 for (pass
= 0; pass
< a
->q
+ 1; pass
++) {
1346 TCGv_i64 tmp
= tcg_temp_new_i64();
1348 read_neon_element64(tmp
, a
->vm
, pass
, MO_64
);
1349 fn(tmp
, cpu_env
, tmp
, constimm
);
1350 write_neon_element64(tmp
, a
->vd
, pass
, MO_64
);
1351 tcg_temp_free_i64(tmp
);
1353 tcg_temp_free_i64(constimm
);
1357 static bool do_2shift_env_32(DisasContext
*s
, arg_2reg_shift
*a
,
1358 NeonGenTwoOpEnvFn
*fn
)
1361 * 2-reg-and-shift operations, size < 3 case, where the
1362 * helper needs to be passed cpu_env.
1364 TCGv_i32 constimm
, tmp
;
1367 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
1371 /* UNDEF accesses to D16-D31 if they don't exist. */
1372 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
1373 ((a
->vd
| a
->vm
) & 0x10)) {
1377 if ((a
->vm
| a
->vd
) & a
->q
) {
1381 if (!vfp_access_check(s
)) {
1386 * To avoid excessive duplication of ops we implement shift
1387 * by immediate using the variable shift operations.
1389 constimm
= tcg_const_i32(dup_const(a
->size
, a
->shift
));
1390 tmp
= tcg_temp_new_i32();
1392 for (pass
= 0; pass
< (a
->q
? 4 : 2); pass
++) {
1393 read_neon_element32(tmp
, a
->vm
, pass
, MO_32
);
1394 fn(tmp
, cpu_env
, tmp
, constimm
);
1395 write_neon_element32(tmp
, a
->vd
, pass
, MO_32
);
1397 tcg_temp_free_i32(tmp
);
1398 tcg_temp_free_i32(constimm
);
1402 #define DO_2SHIFT_ENV(INSN, FUNC) \
1403 static bool trans_##INSN##_64_2sh(DisasContext *s, arg_2reg_shift *a) \
1405 return do_2shift_env_64(s, a, gen_helper_neon_##FUNC##64); \
1407 static bool trans_##INSN##_2sh(DisasContext *s, arg_2reg_shift *a) \
1409 static NeonGenTwoOpEnvFn * const fns[] = { \
1410 gen_helper_neon_##FUNC##8, \
1411 gen_helper_neon_##FUNC##16, \
1412 gen_helper_neon_##FUNC##32, \
1414 assert(a->size < ARRAY_SIZE(fns)); \
1415 return do_2shift_env_32(s, a, fns[a->size]); \
1418 DO_2SHIFT_ENV(VQSHLU
, qshlu_s
)
1419 DO_2SHIFT_ENV(VQSHL_U
, qshl_u
)
1420 DO_2SHIFT_ENV(VQSHL_S
, qshl_s
)
1422 static bool do_2shift_narrow_64(DisasContext
*s
, arg_2reg_shift
*a
,
1423 NeonGenTwo64OpFn
*shiftfn
,
1424 NeonGenNarrowEnvFn
*narrowfn
)
1426 /* 2-reg-and-shift narrowing-shift operations, size == 3 case */
1427 TCGv_i64 constimm
, rm1
, rm2
;
1430 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
1434 /* UNDEF accesses to D16-D31 if they don't exist. */
1435 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
1436 ((a
->vd
| a
->vm
) & 0x10)) {
1444 if (!vfp_access_check(s
)) {
1449 * This is always a right shift, and the shiftfn is always a
1450 * left-shift helper, which thus needs the negated shift count.
1452 constimm
= tcg_const_i64(-a
->shift
);
1453 rm1
= tcg_temp_new_i64();
1454 rm2
= tcg_temp_new_i64();
1455 rd
= tcg_temp_new_i32();
1457 /* Load both inputs first to avoid potential overwrite if rm == rd */
1458 read_neon_element64(rm1
, a
->vm
, 0, MO_64
);
1459 read_neon_element64(rm2
, a
->vm
, 1, MO_64
);
1461 shiftfn(rm1
, rm1
, constimm
);
1462 narrowfn(rd
, cpu_env
, rm1
);
1463 write_neon_element32(rd
, a
->vd
, 0, MO_32
);
1465 shiftfn(rm2
, rm2
, constimm
);
1466 narrowfn(rd
, cpu_env
, rm2
);
1467 write_neon_element32(rd
, a
->vd
, 1, MO_32
);
1469 tcg_temp_free_i32(rd
);
1470 tcg_temp_free_i64(rm1
);
1471 tcg_temp_free_i64(rm2
);
1472 tcg_temp_free_i64(constimm
);
1477 static bool do_2shift_narrow_32(DisasContext
*s
, arg_2reg_shift
*a
,
1478 NeonGenTwoOpFn
*shiftfn
,
1479 NeonGenNarrowEnvFn
*narrowfn
)
1481 /* 2-reg-and-shift narrowing-shift operations, size < 3 case */
1482 TCGv_i32 constimm
, rm1
, rm2
, rm3
, rm4
;
1486 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
1490 /* UNDEF accesses to D16-D31 if they don't exist. */
1491 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
1492 ((a
->vd
| a
->vm
) & 0x10)) {
1500 if (!vfp_access_check(s
)) {
1505 * This is always a right shift, and the shiftfn is always a
1506 * left-shift helper, which thus needs the negated shift count
1507 * duplicated into each lane of the immediate value.
1510 imm
= (uint16_t)(-a
->shift
);
1516 constimm
= tcg_const_i32(imm
);
1518 /* Load all inputs first to avoid potential overwrite */
1519 rm1
= tcg_temp_new_i32();
1520 rm2
= tcg_temp_new_i32();
1521 rm3
= tcg_temp_new_i32();
1522 rm4
= tcg_temp_new_i32();
1523 read_neon_element32(rm1
, a
->vm
, 0, MO_32
);
1524 read_neon_element32(rm2
, a
->vm
, 1, MO_32
);
1525 read_neon_element32(rm3
, a
->vm
, 2, MO_32
);
1526 read_neon_element32(rm4
, a
->vm
, 3, MO_32
);
1527 rtmp
= tcg_temp_new_i64();
1529 shiftfn(rm1
, rm1
, constimm
);
1530 shiftfn(rm2
, rm2
, constimm
);
1532 tcg_gen_concat_i32_i64(rtmp
, rm1
, rm2
);
1533 tcg_temp_free_i32(rm2
);
1535 narrowfn(rm1
, cpu_env
, rtmp
);
1536 write_neon_element32(rm1
, a
->vd
, 0, MO_32
);
1537 tcg_temp_free_i32(rm1
);
1539 shiftfn(rm3
, rm3
, constimm
);
1540 shiftfn(rm4
, rm4
, constimm
);
1541 tcg_temp_free_i32(constimm
);
1543 tcg_gen_concat_i32_i64(rtmp
, rm3
, rm4
);
1544 tcg_temp_free_i32(rm4
);
1546 narrowfn(rm3
, cpu_env
, rtmp
);
1547 tcg_temp_free_i64(rtmp
);
1548 write_neon_element32(rm3
, a
->vd
, 1, MO_32
);
1549 tcg_temp_free_i32(rm3
);
1553 #define DO_2SN_64(INSN, FUNC, NARROWFUNC) \
1554 static bool trans_##INSN##_2sh(DisasContext *s, arg_2reg_shift *a) \
1556 return do_2shift_narrow_64(s, a, FUNC, NARROWFUNC); \
1558 #define DO_2SN_32(INSN, FUNC, NARROWFUNC) \
1559 static bool trans_##INSN##_2sh(DisasContext *s, arg_2reg_shift *a) \
1561 return do_2shift_narrow_32(s, a, FUNC, NARROWFUNC); \
1564 static void gen_neon_narrow_u32(TCGv_i32 dest
, TCGv_ptr env
, TCGv_i64 src
)
1566 tcg_gen_extrl_i64_i32(dest
, src
);
1569 static void gen_neon_narrow_u16(TCGv_i32 dest
, TCGv_ptr env
, TCGv_i64 src
)
1571 gen_helper_neon_narrow_u16(dest
, src
);
1574 static void gen_neon_narrow_u8(TCGv_i32 dest
, TCGv_ptr env
, TCGv_i64 src
)
1576 gen_helper_neon_narrow_u8(dest
, src
);
1579 DO_2SN_64(VSHRN_64
, gen_ushl_i64
, gen_neon_narrow_u32
)
1580 DO_2SN_32(VSHRN_32
, gen_ushl_i32
, gen_neon_narrow_u16
)
1581 DO_2SN_32(VSHRN_16
, gen_helper_neon_shl_u16
, gen_neon_narrow_u8
)
1583 DO_2SN_64(VRSHRN_64
, gen_helper_neon_rshl_u64
, gen_neon_narrow_u32
)
1584 DO_2SN_32(VRSHRN_32
, gen_helper_neon_rshl_u32
, gen_neon_narrow_u16
)
1585 DO_2SN_32(VRSHRN_16
, gen_helper_neon_rshl_u16
, gen_neon_narrow_u8
)
1587 DO_2SN_64(VQSHRUN_64
, gen_sshl_i64
, gen_helper_neon_unarrow_sat32
)
1588 DO_2SN_32(VQSHRUN_32
, gen_sshl_i32
, gen_helper_neon_unarrow_sat16
)
1589 DO_2SN_32(VQSHRUN_16
, gen_helper_neon_shl_s16
, gen_helper_neon_unarrow_sat8
)
1591 DO_2SN_64(VQRSHRUN_64
, gen_helper_neon_rshl_s64
, gen_helper_neon_unarrow_sat32
)
1592 DO_2SN_32(VQRSHRUN_32
, gen_helper_neon_rshl_s32
, gen_helper_neon_unarrow_sat16
)
1593 DO_2SN_32(VQRSHRUN_16
, gen_helper_neon_rshl_s16
, gen_helper_neon_unarrow_sat8
)
1594 DO_2SN_64(VQSHRN_S64
, gen_sshl_i64
, gen_helper_neon_narrow_sat_s32
)
1595 DO_2SN_32(VQSHRN_S32
, gen_sshl_i32
, gen_helper_neon_narrow_sat_s16
)
1596 DO_2SN_32(VQSHRN_S16
, gen_helper_neon_shl_s16
, gen_helper_neon_narrow_sat_s8
)
1598 DO_2SN_64(VQRSHRN_S64
, gen_helper_neon_rshl_s64
, gen_helper_neon_narrow_sat_s32
)
1599 DO_2SN_32(VQRSHRN_S32
, gen_helper_neon_rshl_s32
, gen_helper_neon_narrow_sat_s16
)
1600 DO_2SN_32(VQRSHRN_S16
, gen_helper_neon_rshl_s16
, gen_helper_neon_narrow_sat_s8
)
1602 DO_2SN_64(VQSHRN_U64
, gen_ushl_i64
, gen_helper_neon_narrow_sat_u32
)
1603 DO_2SN_32(VQSHRN_U32
, gen_ushl_i32
, gen_helper_neon_narrow_sat_u16
)
1604 DO_2SN_32(VQSHRN_U16
, gen_helper_neon_shl_u16
, gen_helper_neon_narrow_sat_u8
)
1606 DO_2SN_64(VQRSHRN_U64
, gen_helper_neon_rshl_u64
, gen_helper_neon_narrow_sat_u32
)
1607 DO_2SN_32(VQRSHRN_U32
, gen_helper_neon_rshl_u32
, gen_helper_neon_narrow_sat_u16
)
1608 DO_2SN_32(VQRSHRN_U16
, gen_helper_neon_rshl_u16
, gen_helper_neon_narrow_sat_u8
)
1610 static bool do_vshll_2sh(DisasContext
*s
, arg_2reg_shift
*a
,
1611 NeonGenWidenFn
*widenfn
, bool u
)
1615 uint64_t widen_mask
= 0;
1617 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
1621 /* UNDEF accesses to D16-D31 if they don't exist. */
1622 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
1623 ((a
->vd
| a
->vm
) & 0x10)) {
1631 if (!vfp_access_check(s
)) {
1636 * This is a widen-and-shift operation. The shift is always less
1637 * than the width of the source type, so after widening the input
1638 * vector we can simply shift the whole 64-bit widened register,
1639 * and then clear the potential overflow bits resulting from left
1640 * bits of the narrow input appearing as right bits of the left
1641 * neighbour narrow input. Calculate a mask of bits to clear.
1643 if ((a
->shift
!= 0) && (a
->size
< 2 || u
)) {
1644 int esize
= 8 << a
->size
;
1645 widen_mask
= MAKE_64BIT_MASK(0, esize
);
1646 widen_mask
>>= esize
- a
->shift
;
1647 widen_mask
= dup_const(a
->size
+ 1, widen_mask
);
1650 rm0
= tcg_temp_new_i32();
1651 rm1
= tcg_temp_new_i32();
1652 read_neon_element32(rm0
, a
->vm
, 0, MO_32
);
1653 read_neon_element32(rm1
, a
->vm
, 1, MO_32
);
1654 tmp
= tcg_temp_new_i64();
1657 tcg_temp_free_i32(rm0
);
1658 if (a
->shift
!= 0) {
1659 tcg_gen_shli_i64(tmp
, tmp
, a
->shift
);
1660 tcg_gen_andi_i64(tmp
, tmp
, ~widen_mask
);
1662 write_neon_element64(tmp
, a
->vd
, 0, MO_64
);
1665 tcg_temp_free_i32(rm1
);
1666 if (a
->shift
!= 0) {
1667 tcg_gen_shli_i64(tmp
, tmp
, a
->shift
);
1668 tcg_gen_andi_i64(tmp
, tmp
, ~widen_mask
);
1670 write_neon_element64(tmp
, a
->vd
, 1, MO_64
);
1671 tcg_temp_free_i64(tmp
);
1675 static bool trans_VSHLL_S_2sh(DisasContext
*s
, arg_2reg_shift
*a
)
1677 static NeonGenWidenFn
* const widenfn
[] = {
1678 gen_helper_neon_widen_s8
,
1679 gen_helper_neon_widen_s16
,
1680 tcg_gen_ext_i32_i64
,
1682 return do_vshll_2sh(s
, a
, widenfn
[a
->size
], false);
1685 static bool trans_VSHLL_U_2sh(DisasContext
*s
, arg_2reg_shift
*a
)
1687 static NeonGenWidenFn
* const widenfn
[] = {
1688 gen_helper_neon_widen_u8
,
1689 gen_helper_neon_widen_u16
,
1690 tcg_gen_extu_i32_i64
,
1692 return do_vshll_2sh(s
, a
, widenfn
[a
->size
], true);
1695 static bool do_fp_2sh(DisasContext
*s
, arg_2reg_shift
*a
,
1696 gen_helper_gvec_2_ptr
*fn
)
1698 /* FP operations in 2-reg-and-shift group */
1699 int vec_size
= a
->q
? 16 : 8;
1700 int rd_ofs
= neon_full_reg_offset(a
->vd
);
1701 int rm_ofs
= neon_full_reg_offset(a
->vm
);
1704 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
1708 if (a
->size
== MO_16
) {
1709 if (!dc_isar_feature(aa32_fp16_arith
, s
)) {
1714 /* UNDEF accesses to D16-D31 if they don't exist. */
1715 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
1716 ((a
->vd
| a
->vm
) & 0x10)) {
1720 if ((a
->vm
| a
->vd
) & a
->q
) {
1724 if (!vfp_access_check(s
)) {
1728 fpst
= fpstatus_ptr(a
->size
== MO_16
? FPST_STD_F16
: FPST_STD
);
1729 tcg_gen_gvec_2_ptr(rd_ofs
, rm_ofs
, fpst
, vec_size
, vec_size
, a
->shift
, fn
);
1730 tcg_temp_free_ptr(fpst
);
1734 #define DO_FP_2SH(INSN, FUNC) \
1735 static bool trans_##INSN##_2sh(DisasContext *s, arg_2reg_shift *a) \
1737 return do_fp_2sh(s, a, FUNC); \
1740 DO_FP_2SH(VCVT_SF
, gen_helper_gvec_vcvt_sf
)
1741 DO_FP_2SH(VCVT_UF
, gen_helper_gvec_vcvt_uf
)
1742 DO_FP_2SH(VCVT_FS
, gen_helper_gvec_vcvt_fs
)
1743 DO_FP_2SH(VCVT_FU
, gen_helper_gvec_vcvt_fu
)
1745 DO_FP_2SH(VCVT_SH
, gen_helper_gvec_vcvt_sh
)
1746 DO_FP_2SH(VCVT_UH
, gen_helper_gvec_vcvt_uh
)
1747 DO_FP_2SH(VCVT_HS
, gen_helper_gvec_vcvt_hs
)
1748 DO_FP_2SH(VCVT_HU
, gen_helper_gvec_vcvt_hu
)
1750 static uint64_t asimd_imm_const(uint32_t imm
, int cmode
, int op
)
1753 * Expand the encoded constant.
1754 * Note that cmode = 2,3,4,5,6,7,10,11,12,13 imm=0 is UNPREDICTABLE.
1755 * We choose to not special-case this and will behave as if a
1756 * valid constant encoding of 0 had been given.
1757 * cmode = 15 op = 1 must UNDEF; we assume decode has handled that.
1776 imm
= (imm
<< 8) | (imm
<< 24);
1779 imm
= (imm
<< 8) | 0xff;
1782 imm
= (imm
<< 16) | 0xffff;
1787 * This is the only case where the top and bottom 32 bits
1788 * of the encoded constant differ.
1793 for (n
= 0; n
< 8; n
++) {
1794 if (imm
& (1 << n
)) {
1795 imm64
|= (0xffULL
<< (n
* 8));
1800 imm
|= (imm
<< 8) | (imm
<< 16) | (imm
<< 24);
1803 imm
= ((imm
& 0x80) << 24) | ((imm
& 0x3f) << 19)
1804 | ((imm
& 0x40) ? (0x1f << 25) : (1 << 30));
1810 return dup_const(MO_32
, imm
);
1813 static bool do_1reg_imm(DisasContext
*s
, arg_1reg_imm
*a
,
1817 int reg_ofs
, vec_size
;
1819 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
1823 /* UNDEF accesses to D16-D31 if they don't exist. */
1824 if (!dc_isar_feature(aa32_simd_r32
, s
) && (a
->vd
& 0x10)) {
1832 if (!vfp_access_check(s
)) {
1836 reg_ofs
= neon_full_reg_offset(a
->vd
);
1837 vec_size
= a
->q
? 16 : 8;
1838 imm
= asimd_imm_const(a
->imm
, a
->cmode
, a
->op
);
1840 fn(MO_64
, reg_ofs
, reg_ofs
, imm
, vec_size
, vec_size
);
1844 static void gen_VMOV_1r(unsigned vece
, uint32_t dofs
, uint32_t aofs
,
1845 int64_t c
, uint32_t oprsz
, uint32_t maxsz
)
1847 tcg_gen_gvec_dup_imm(MO_64
, dofs
, oprsz
, maxsz
, c
);
1850 static bool trans_Vimm_1r(DisasContext
*s
, arg_1reg_imm
*a
)
1852 /* Handle decode of cmode/op here between VORR/VBIC/VMOV */
1855 if ((a
->cmode
& 1) && a
->cmode
< 12) {
1856 /* for op=1, the imm will be inverted, so BIC becomes AND. */
1857 fn
= a
->op
? tcg_gen_gvec_andi
: tcg_gen_gvec_ori
;
1859 /* There is one unallocated cmode/op combination in this space */
1860 if (a
->cmode
== 15 && a
->op
== 1) {
1865 return do_1reg_imm(s
, a
, fn
);
1868 static bool do_prewiden_3d(DisasContext
*s
, arg_3diff
*a
,
1869 NeonGenWidenFn
*widenfn
,
1870 NeonGenTwo64OpFn
*opfn
,
1871 int src1_mop
, int src2_mop
)
1873 /* 3-regs different lengths, prewidening case (VADDL/VSUBL/VAADW/VSUBW) */
1874 TCGv_i64 rn0_64
, rn1_64
, rm_64
;
1876 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
1880 /* UNDEF accesses to D16-D31 if they don't exist. */
1881 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
1882 ((a
->vd
| a
->vn
| a
->vm
) & 0x10)) {
1887 /* size == 3 case, which is an entirely different insn group */
1891 if ((a
->vd
& 1) || (src1_mop
== MO_Q
&& (a
->vn
& 1))) {
1895 if (!vfp_access_check(s
)) {
1899 rn0_64
= tcg_temp_new_i64();
1900 rn1_64
= tcg_temp_new_i64();
1901 rm_64
= tcg_temp_new_i64();
1903 if (src1_mop
>= 0) {
1904 read_neon_element64(rn0_64
, a
->vn
, 0, src1_mop
);
1906 TCGv_i32 tmp
= tcg_temp_new_i32();
1907 read_neon_element32(tmp
, a
->vn
, 0, MO_32
);
1908 widenfn(rn0_64
, tmp
);
1909 tcg_temp_free_i32(tmp
);
1911 if (src2_mop
>= 0) {
1912 read_neon_element64(rm_64
, a
->vm
, 0, src2_mop
);
1914 TCGv_i32 tmp
= tcg_temp_new_i32();
1915 read_neon_element32(tmp
, a
->vm
, 0, MO_32
);
1916 widenfn(rm_64
, tmp
);
1917 tcg_temp_free_i32(tmp
);
1920 opfn(rn0_64
, rn0_64
, rm_64
);
1923 * Load second pass inputs before storing the first pass result, to
1924 * avoid incorrect results if a narrow input overlaps with the result.
1926 if (src1_mop
>= 0) {
1927 read_neon_element64(rn1_64
, a
->vn
, 1, src1_mop
);
1929 TCGv_i32 tmp
= tcg_temp_new_i32();
1930 read_neon_element32(tmp
, a
->vn
, 1, MO_32
);
1931 widenfn(rn1_64
, tmp
);
1932 tcg_temp_free_i32(tmp
);
1934 if (src2_mop
>= 0) {
1935 read_neon_element64(rm_64
, a
->vm
, 1, src2_mop
);
1937 TCGv_i32 tmp
= tcg_temp_new_i32();
1938 read_neon_element32(tmp
, a
->vm
, 1, MO_32
);
1939 widenfn(rm_64
, tmp
);
1940 tcg_temp_free_i32(tmp
);
1943 write_neon_element64(rn0_64
, a
->vd
, 0, MO_64
);
1945 opfn(rn1_64
, rn1_64
, rm_64
);
1946 write_neon_element64(rn1_64
, a
->vd
, 1, MO_64
);
1948 tcg_temp_free_i64(rn0_64
);
1949 tcg_temp_free_i64(rn1_64
);
1950 tcg_temp_free_i64(rm_64
);
1955 #define DO_PREWIDEN(INSN, S, OP, SRC1WIDE, SIGN) \
1956 static bool trans_##INSN##_3d(DisasContext *s, arg_3diff *a) \
1958 static NeonGenWidenFn * const widenfn[] = { \
1959 gen_helper_neon_widen_##S##8, \
1960 gen_helper_neon_widen_##S##16, \
1963 static NeonGenTwo64OpFn * const addfn[] = { \
1964 gen_helper_neon_##OP##l_u16, \
1965 gen_helper_neon_##OP##l_u32, \
1966 tcg_gen_##OP##_i64, \
1969 int narrow_mop = a->size == MO_32 ? MO_32 | SIGN : -1; \
1970 return do_prewiden_3d(s, a, widenfn[a->size], addfn[a->size], \
1971 SRC1WIDE ? MO_Q : narrow_mop, \
1975 DO_PREWIDEN(VADDL_S
, s
, add
, false, MO_SIGN
)
1976 DO_PREWIDEN(VADDL_U
, u
, add
, false, 0)
1977 DO_PREWIDEN(VSUBL_S
, s
, sub
, false, MO_SIGN
)
1978 DO_PREWIDEN(VSUBL_U
, u
, sub
, false, 0)
1979 DO_PREWIDEN(VADDW_S
, s
, add
, true, MO_SIGN
)
1980 DO_PREWIDEN(VADDW_U
, u
, add
, true, 0)
1981 DO_PREWIDEN(VSUBW_S
, s
, sub
, true, MO_SIGN
)
1982 DO_PREWIDEN(VSUBW_U
, u
, sub
, true, 0)
1984 static bool do_narrow_3d(DisasContext
*s
, arg_3diff
*a
,
1985 NeonGenTwo64OpFn
*opfn
, NeonGenNarrowFn
*narrowfn
)
1987 /* 3-regs different lengths, narrowing (VADDHN/VSUBHN/VRADDHN/VRSUBHN) */
1988 TCGv_i64 rn_64
, rm_64
;
1991 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
1995 /* UNDEF accesses to D16-D31 if they don't exist. */
1996 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
1997 ((a
->vd
| a
->vn
| a
->vm
) & 0x10)) {
2001 if (!opfn
|| !narrowfn
) {
2002 /* size == 3 case, which is an entirely different insn group */
2006 if ((a
->vn
| a
->vm
) & 1) {
2010 if (!vfp_access_check(s
)) {
2014 rn_64
= tcg_temp_new_i64();
2015 rm_64
= tcg_temp_new_i64();
2016 rd0
= tcg_temp_new_i32();
2017 rd1
= tcg_temp_new_i32();
2019 read_neon_element64(rn_64
, a
->vn
, 0, MO_64
);
2020 read_neon_element64(rm_64
, a
->vm
, 0, MO_64
);
2022 opfn(rn_64
, rn_64
, rm_64
);
2024 narrowfn(rd0
, rn_64
);
2026 read_neon_element64(rn_64
, a
->vn
, 1, MO_64
);
2027 read_neon_element64(rm_64
, a
->vm
, 1, MO_64
);
2029 opfn(rn_64
, rn_64
, rm_64
);
2031 narrowfn(rd1
, rn_64
);
2033 write_neon_element32(rd0
, a
->vd
, 0, MO_32
);
2034 write_neon_element32(rd1
, a
->vd
, 1, MO_32
);
2036 tcg_temp_free_i32(rd0
);
2037 tcg_temp_free_i32(rd1
);
2038 tcg_temp_free_i64(rn_64
);
2039 tcg_temp_free_i64(rm_64
);
2044 #define DO_NARROW_3D(INSN, OP, NARROWTYPE, EXTOP) \
2045 static bool trans_##INSN##_3d(DisasContext *s, arg_3diff *a) \
2047 static NeonGenTwo64OpFn * const addfn[] = { \
2048 gen_helper_neon_##OP##l_u16, \
2049 gen_helper_neon_##OP##l_u32, \
2050 tcg_gen_##OP##_i64, \
2053 static NeonGenNarrowFn * const narrowfn[] = { \
2054 gen_helper_neon_##NARROWTYPE##_high_u8, \
2055 gen_helper_neon_##NARROWTYPE##_high_u16, \
2059 return do_narrow_3d(s, a, addfn[a->size], narrowfn[a->size]); \
2062 static void gen_narrow_round_high_u32(TCGv_i32 rd
, TCGv_i64 rn
)
2064 tcg_gen_addi_i64(rn
, rn
, 1u << 31);
2065 tcg_gen_extrh_i64_i32(rd
, rn
);
2068 DO_NARROW_3D(VADDHN
, add
, narrow
, tcg_gen_extrh_i64_i32
)
2069 DO_NARROW_3D(VSUBHN
, sub
, narrow
, tcg_gen_extrh_i64_i32
)
2070 DO_NARROW_3D(VRADDHN
, add
, narrow_round
, gen_narrow_round_high_u32
)
2071 DO_NARROW_3D(VRSUBHN
, sub
, narrow_round
, gen_narrow_round_high_u32
)
2073 static bool do_long_3d(DisasContext
*s
, arg_3diff
*a
,
2074 NeonGenTwoOpWidenFn
*opfn
,
2075 NeonGenTwo64OpFn
*accfn
)
2078 * 3-regs different lengths, long operations.
2079 * These perform an operation on two inputs that returns a double-width
2080 * result, and then possibly perform an accumulation operation of
2081 * that result into the double-width destination.
2083 TCGv_i64 rd0
, rd1
, tmp
;
2086 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
2090 /* UNDEF accesses to D16-D31 if they don't exist. */
2091 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
2092 ((a
->vd
| a
->vn
| a
->vm
) & 0x10)) {
2097 /* size == 3 case, which is an entirely different insn group */
2105 if (!vfp_access_check(s
)) {
2109 rd0
= tcg_temp_new_i64();
2110 rd1
= tcg_temp_new_i64();
2112 rn
= tcg_temp_new_i32();
2113 rm
= tcg_temp_new_i32();
2114 read_neon_element32(rn
, a
->vn
, 0, MO_32
);
2115 read_neon_element32(rm
, a
->vm
, 0, MO_32
);
2118 read_neon_element32(rn
, a
->vn
, 1, MO_32
);
2119 read_neon_element32(rm
, a
->vm
, 1, MO_32
);
2121 tcg_temp_free_i32(rn
);
2122 tcg_temp_free_i32(rm
);
2124 /* Don't store results until after all loads: they might overlap */
2126 tmp
= tcg_temp_new_i64();
2127 read_neon_element64(tmp
, a
->vd
, 0, MO_64
);
2128 accfn(rd0
, tmp
, rd0
);
2129 read_neon_element64(tmp
, a
->vd
, 1, MO_64
);
2130 accfn(rd1
, tmp
, rd1
);
2131 tcg_temp_free_i64(tmp
);
2134 write_neon_element64(rd0
, a
->vd
, 0, MO_64
);
2135 write_neon_element64(rd1
, a
->vd
, 1, MO_64
);
2136 tcg_temp_free_i64(rd0
);
2137 tcg_temp_free_i64(rd1
);
2142 static bool trans_VABDL_S_3d(DisasContext
*s
, arg_3diff
*a
)
2144 static NeonGenTwoOpWidenFn
* const opfn
[] = {
2145 gen_helper_neon_abdl_s16
,
2146 gen_helper_neon_abdl_s32
,
2147 gen_helper_neon_abdl_s64
,
2151 return do_long_3d(s
, a
, opfn
[a
->size
], NULL
);
2154 static bool trans_VABDL_U_3d(DisasContext
*s
, arg_3diff
*a
)
2156 static NeonGenTwoOpWidenFn
* const opfn
[] = {
2157 gen_helper_neon_abdl_u16
,
2158 gen_helper_neon_abdl_u32
,
2159 gen_helper_neon_abdl_u64
,
2163 return do_long_3d(s
, a
, opfn
[a
->size
], NULL
);
2166 static bool trans_VABAL_S_3d(DisasContext
*s
, arg_3diff
*a
)
2168 static NeonGenTwoOpWidenFn
* const opfn
[] = {
2169 gen_helper_neon_abdl_s16
,
2170 gen_helper_neon_abdl_s32
,
2171 gen_helper_neon_abdl_s64
,
2174 static NeonGenTwo64OpFn
* const addfn
[] = {
2175 gen_helper_neon_addl_u16
,
2176 gen_helper_neon_addl_u32
,
2181 return do_long_3d(s
, a
, opfn
[a
->size
], addfn
[a
->size
]);
2184 static bool trans_VABAL_U_3d(DisasContext
*s
, arg_3diff
*a
)
2186 static NeonGenTwoOpWidenFn
* const opfn
[] = {
2187 gen_helper_neon_abdl_u16
,
2188 gen_helper_neon_abdl_u32
,
2189 gen_helper_neon_abdl_u64
,
2192 static NeonGenTwo64OpFn
* const addfn
[] = {
2193 gen_helper_neon_addl_u16
,
2194 gen_helper_neon_addl_u32
,
2199 return do_long_3d(s
, a
, opfn
[a
->size
], addfn
[a
->size
]);
2202 static void gen_mull_s32(TCGv_i64 rd
, TCGv_i32 rn
, TCGv_i32 rm
)
2204 TCGv_i32 lo
= tcg_temp_new_i32();
2205 TCGv_i32 hi
= tcg_temp_new_i32();
2207 tcg_gen_muls2_i32(lo
, hi
, rn
, rm
);
2208 tcg_gen_concat_i32_i64(rd
, lo
, hi
);
2210 tcg_temp_free_i32(lo
);
2211 tcg_temp_free_i32(hi
);
2214 static void gen_mull_u32(TCGv_i64 rd
, TCGv_i32 rn
, TCGv_i32 rm
)
2216 TCGv_i32 lo
= tcg_temp_new_i32();
2217 TCGv_i32 hi
= tcg_temp_new_i32();
2219 tcg_gen_mulu2_i32(lo
, hi
, rn
, rm
);
2220 tcg_gen_concat_i32_i64(rd
, lo
, hi
);
2222 tcg_temp_free_i32(lo
);
2223 tcg_temp_free_i32(hi
);
2226 static bool trans_VMULL_S_3d(DisasContext
*s
, arg_3diff
*a
)
2228 static NeonGenTwoOpWidenFn
* const opfn
[] = {
2229 gen_helper_neon_mull_s8
,
2230 gen_helper_neon_mull_s16
,
2235 return do_long_3d(s
, a
, opfn
[a
->size
], NULL
);
2238 static bool trans_VMULL_U_3d(DisasContext
*s
, arg_3diff
*a
)
2240 static NeonGenTwoOpWidenFn
* const opfn
[] = {
2241 gen_helper_neon_mull_u8
,
2242 gen_helper_neon_mull_u16
,
2247 return do_long_3d(s
, a
, opfn
[a
->size
], NULL
);
2250 #define DO_VMLAL(INSN,MULL,ACC) \
2251 static bool trans_##INSN##_3d(DisasContext *s, arg_3diff *a) \
2253 static NeonGenTwoOpWidenFn * const opfn[] = { \
2254 gen_helper_neon_##MULL##8, \
2255 gen_helper_neon_##MULL##16, \
2259 static NeonGenTwo64OpFn * const accfn[] = { \
2260 gen_helper_neon_##ACC##l_u16, \
2261 gen_helper_neon_##ACC##l_u32, \
2262 tcg_gen_##ACC##_i64, \
2265 return do_long_3d(s, a, opfn[a->size], accfn[a->size]); \
2268 DO_VMLAL(VMLAL_S
,mull_s
,add
)
2269 DO_VMLAL(VMLAL_U
,mull_u
,add
)
2270 DO_VMLAL(VMLSL_S
,mull_s
,sub
)
2271 DO_VMLAL(VMLSL_U
,mull_u
,sub
)
2273 static void gen_VQDMULL_16(TCGv_i64 rd
, TCGv_i32 rn
, TCGv_i32 rm
)
2275 gen_helper_neon_mull_s16(rd
, rn
, rm
);
2276 gen_helper_neon_addl_saturate_s32(rd
, cpu_env
, rd
, rd
);
2279 static void gen_VQDMULL_32(TCGv_i64 rd
, TCGv_i32 rn
, TCGv_i32 rm
)
2281 gen_mull_s32(rd
, rn
, rm
);
2282 gen_helper_neon_addl_saturate_s64(rd
, cpu_env
, rd
, rd
);
2285 static bool trans_VQDMULL_3d(DisasContext
*s
, arg_3diff
*a
)
2287 static NeonGenTwoOpWidenFn
* const opfn
[] = {
2294 return do_long_3d(s
, a
, opfn
[a
->size
], NULL
);
2297 static void gen_VQDMLAL_acc_16(TCGv_i64 rd
, TCGv_i64 rn
, TCGv_i64 rm
)
2299 gen_helper_neon_addl_saturate_s32(rd
, cpu_env
, rn
, rm
);
2302 static void gen_VQDMLAL_acc_32(TCGv_i64 rd
, TCGv_i64 rn
, TCGv_i64 rm
)
2304 gen_helper_neon_addl_saturate_s64(rd
, cpu_env
, rn
, rm
);
2307 static bool trans_VQDMLAL_3d(DisasContext
*s
, arg_3diff
*a
)
2309 static NeonGenTwoOpWidenFn
* const opfn
[] = {
2315 static NeonGenTwo64OpFn
* const accfn
[] = {
2322 return do_long_3d(s
, a
, opfn
[a
->size
], accfn
[a
->size
]);
2325 static void gen_VQDMLSL_acc_16(TCGv_i64 rd
, TCGv_i64 rn
, TCGv_i64 rm
)
2327 gen_helper_neon_negl_u32(rm
, rm
);
2328 gen_helper_neon_addl_saturate_s32(rd
, cpu_env
, rn
, rm
);
2331 static void gen_VQDMLSL_acc_32(TCGv_i64 rd
, TCGv_i64 rn
, TCGv_i64 rm
)
2333 tcg_gen_neg_i64(rm
, rm
);
2334 gen_helper_neon_addl_saturate_s64(rd
, cpu_env
, rn
, rm
);
2337 static bool trans_VQDMLSL_3d(DisasContext
*s
, arg_3diff
*a
)
2339 static NeonGenTwoOpWidenFn
* const opfn
[] = {
2345 static NeonGenTwo64OpFn
* const accfn
[] = {
2352 return do_long_3d(s
, a
, opfn
[a
->size
], accfn
[a
->size
]);
2355 static bool trans_VMULL_P_3d(DisasContext
*s
, arg_3diff
*a
)
2357 gen_helper_gvec_3
*fn_gvec
;
2359 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
2363 /* UNDEF accesses to D16-D31 if they don't exist. */
2364 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
2365 ((a
->vd
| a
->vn
| a
->vm
) & 0x10)) {
2375 fn_gvec
= gen_helper_neon_pmull_h
;
2378 if (!dc_isar_feature(aa32_pmull
, s
)) {
2381 fn_gvec
= gen_helper_gvec_pmull_q
;
2387 if (!vfp_access_check(s
)) {
2391 tcg_gen_gvec_3_ool(neon_full_reg_offset(a
->vd
),
2392 neon_full_reg_offset(a
->vn
),
2393 neon_full_reg_offset(a
->vm
),
2394 16, 16, 0, fn_gvec
);
2398 static void gen_neon_dup_low16(TCGv_i32 var
)
2400 TCGv_i32 tmp
= tcg_temp_new_i32();
2401 tcg_gen_ext16u_i32(var
, var
);
2402 tcg_gen_shli_i32(tmp
, var
, 16);
2403 tcg_gen_or_i32(var
, var
, tmp
);
2404 tcg_temp_free_i32(tmp
);
2407 static void gen_neon_dup_high16(TCGv_i32 var
)
2409 TCGv_i32 tmp
= tcg_temp_new_i32();
2410 tcg_gen_andi_i32(var
, var
, 0xffff0000);
2411 tcg_gen_shri_i32(tmp
, var
, 16);
2412 tcg_gen_or_i32(var
, var
, tmp
);
2413 tcg_temp_free_i32(tmp
);
2416 static inline TCGv_i32
neon_get_scalar(int size
, int reg
)
2418 TCGv_i32 tmp
= tcg_temp_new_i32();
2419 if (size
== MO_16
) {
2420 read_neon_element32(tmp
, reg
& 7, reg
>> 4, MO_32
);
2422 gen_neon_dup_high16(tmp
);
2424 gen_neon_dup_low16(tmp
);
2427 read_neon_element32(tmp
, reg
& 15, reg
>> 4, MO_32
);
2432 static bool do_2scalar(DisasContext
*s
, arg_2scalar
*a
,
2433 NeonGenTwoOpFn
*opfn
, NeonGenTwoOpFn
*accfn
)
2436 * Two registers and a scalar: perform an operation between
2437 * the input elements and the scalar, and then possibly
2438 * perform an accumulation operation of that result into the
2441 TCGv_i32 scalar
, tmp
;
2444 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
2448 /* UNDEF accesses to D16-D31 if they don't exist. */
2449 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
2450 ((a
->vd
| a
->vn
| a
->vm
) & 0x10)) {
2455 /* Bad size (including size == 3, which is a different insn group) */
2459 if (a
->q
&& ((a
->vd
| a
->vn
) & 1)) {
2463 if (!vfp_access_check(s
)) {
2467 scalar
= neon_get_scalar(a
->size
, a
->vm
);
2468 tmp
= tcg_temp_new_i32();
2470 for (pass
= 0; pass
< (a
->q
? 4 : 2); pass
++) {
2471 read_neon_element32(tmp
, a
->vn
, pass
, MO_32
);
2472 opfn(tmp
, tmp
, scalar
);
2474 TCGv_i32 rd
= tcg_temp_new_i32();
2475 read_neon_element32(rd
, a
->vd
, pass
, MO_32
);
2476 accfn(tmp
, rd
, tmp
);
2477 tcg_temp_free_i32(rd
);
2479 write_neon_element32(tmp
, a
->vd
, pass
, MO_32
);
2481 tcg_temp_free_i32(tmp
);
2482 tcg_temp_free_i32(scalar
);
2486 static bool trans_VMUL_2sc(DisasContext
*s
, arg_2scalar
*a
)
2488 static NeonGenTwoOpFn
* const opfn
[] = {
2490 gen_helper_neon_mul_u16
,
2495 return do_2scalar(s
, a
, opfn
[a
->size
], NULL
);
2498 static bool trans_VMLA_2sc(DisasContext
*s
, arg_2scalar
*a
)
2500 static NeonGenTwoOpFn
* const opfn
[] = {
2502 gen_helper_neon_mul_u16
,
2506 static NeonGenTwoOpFn
* const accfn
[] = {
2508 gen_helper_neon_add_u16
,
2513 return do_2scalar(s
, a
, opfn
[a
->size
], accfn
[a
->size
]);
2516 static bool trans_VMLS_2sc(DisasContext
*s
, arg_2scalar
*a
)
2518 static NeonGenTwoOpFn
* const opfn
[] = {
2520 gen_helper_neon_mul_u16
,
2524 static NeonGenTwoOpFn
* const accfn
[] = {
2526 gen_helper_neon_sub_u16
,
2531 return do_2scalar(s
, a
, opfn
[a
->size
], accfn
[a
->size
]);
2534 static bool do_2scalar_fp_vec(DisasContext
*s
, arg_2scalar
*a
,
2535 gen_helper_gvec_3_ptr
*fn
)
2537 /* Two registers and a scalar, using gvec */
2538 int vec_size
= a
->q
? 16 : 8;
2539 int rd_ofs
= neon_full_reg_offset(a
->vd
);
2540 int rn_ofs
= neon_full_reg_offset(a
->vn
);
2545 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
2549 /* UNDEF accesses to D16-D31 if they don't exist. */
2550 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
2551 ((a
->vd
| a
->vn
| a
->vm
) & 0x10)) {
2556 /* Bad size (including size == 3, which is a different insn group) */
2560 if (a
->q
&& ((a
->vd
| a
->vn
) & 1)) {
2564 if (!vfp_access_check(s
)) {
2568 /* a->vm is M:Vm, which encodes both register and index */
2569 idx
= extract32(a
->vm
, a
->size
+ 2, 2);
2570 a
->vm
= extract32(a
->vm
, 0, a
->size
+ 2);
2571 rm_ofs
= neon_full_reg_offset(a
->vm
);
2573 fpstatus
= fpstatus_ptr(a
->size
== 1 ? FPST_STD_F16
: FPST_STD
);
2574 tcg_gen_gvec_3_ptr(rd_ofs
, rn_ofs
, rm_ofs
, fpstatus
,
2575 vec_size
, vec_size
, idx
, fn
);
2576 tcg_temp_free_ptr(fpstatus
);
2580 #define DO_VMUL_F_2sc(NAME, FUNC) \
2581 static bool trans_##NAME##_F_2sc(DisasContext *s, arg_2scalar *a) \
2583 static gen_helper_gvec_3_ptr * const opfn[] = { \
2585 gen_helper_##FUNC##_h, \
2586 gen_helper_##FUNC##_s, \
2589 if (a->size == MO_16 && !dc_isar_feature(aa32_fp16_arith, s)) { \
2592 return do_2scalar_fp_vec(s, a, opfn[a->size]); \
2595 DO_VMUL_F_2sc(VMUL
, gvec_fmul_idx
)
2596 DO_VMUL_F_2sc(VMLA
, gvec_fmla_nf_idx
)
2597 DO_VMUL_F_2sc(VMLS
, gvec_fmls_nf_idx
)
2599 WRAP_ENV_FN(gen_VQDMULH_16
, gen_helper_neon_qdmulh_s16
)
2600 WRAP_ENV_FN(gen_VQDMULH_32
, gen_helper_neon_qdmulh_s32
)
2601 WRAP_ENV_FN(gen_VQRDMULH_16
, gen_helper_neon_qrdmulh_s16
)
2602 WRAP_ENV_FN(gen_VQRDMULH_32
, gen_helper_neon_qrdmulh_s32
)
2604 static bool trans_VQDMULH_2sc(DisasContext
*s
, arg_2scalar
*a
)
2606 static NeonGenTwoOpFn
* const opfn
[] = {
2613 return do_2scalar(s
, a
, opfn
[a
->size
], NULL
);
2616 static bool trans_VQRDMULH_2sc(DisasContext
*s
, arg_2scalar
*a
)
2618 static NeonGenTwoOpFn
* const opfn
[] = {
2625 return do_2scalar(s
, a
, opfn
[a
->size
], NULL
);
2628 static bool do_vqrdmlah_2sc(DisasContext
*s
, arg_2scalar
*a
,
2629 NeonGenThreeOpEnvFn
*opfn
)
2632 * VQRDMLAH/VQRDMLSH: this is like do_2scalar, but the opfn
2633 * performs a kind of fused op-then-accumulate using a helper
2634 * function that takes all of rd, rn and the scalar at once.
2636 TCGv_i32 scalar
, rn
, rd
;
2639 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
2643 if (!dc_isar_feature(aa32_rdm
, s
)) {
2647 /* UNDEF accesses to D16-D31 if they don't exist. */
2648 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
2649 ((a
->vd
| a
->vn
| a
->vm
) & 0x10)) {
2654 /* Bad size (including size == 3, which is a different insn group) */
2658 if (a
->q
&& ((a
->vd
| a
->vn
) & 1)) {
2662 if (!vfp_access_check(s
)) {
2666 scalar
= neon_get_scalar(a
->size
, a
->vm
);
2667 rn
= tcg_temp_new_i32();
2668 rd
= tcg_temp_new_i32();
2670 for (pass
= 0; pass
< (a
->q
? 4 : 2); pass
++) {
2671 read_neon_element32(rn
, a
->vn
, pass
, MO_32
);
2672 read_neon_element32(rd
, a
->vd
, pass
, MO_32
);
2673 opfn(rd
, cpu_env
, rn
, scalar
, rd
);
2674 write_neon_element32(rd
, a
->vd
, pass
, MO_32
);
2676 tcg_temp_free_i32(rn
);
2677 tcg_temp_free_i32(rd
);
2678 tcg_temp_free_i32(scalar
);
2683 static bool trans_VQRDMLAH_2sc(DisasContext
*s
, arg_2scalar
*a
)
2685 static NeonGenThreeOpEnvFn
*opfn
[] = {
2687 gen_helper_neon_qrdmlah_s16
,
2688 gen_helper_neon_qrdmlah_s32
,
2691 return do_vqrdmlah_2sc(s
, a
, opfn
[a
->size
]);
2694 static bool trans_VQRDMLSH_2sc(DisasContext
*s
, arg_2scalar
*a
)
2696 static NeonGenThreeOpEnvFn
*opfn
[] = {
2698 gen_helper_neon_qrdmlsh_s16
,
2699 gen_helper_neon_qrdmlsh_s32
,
2702 return do_vqrdmlah_2sc(s
, a
, opfn
[a
->size
]);
2705 static bool do_2scalar_long(DisasContext
*s
, arg_2scalar
*a
,
2706 NeonGenTwoOpWidenFn
*opfn
,
2707 NeonGenTwo64OpFn
*accfn
)
2710 * Two registers and a scalar, long operations: perform an
2711 * operation on the input elements and the scalar which produces
2712 * a double-width result, and then possibly perform an accumulation
2713 * operation of that result into the destination.
2715 TCGv_i32 scalar
, rn
;
2716 TCGv_i64 rn0_64
, rn1_64
;
2718 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
2722 /* UNDEF accesses to D16-D31 if they don't exist. */
2723 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
2724 ((a
->vd
| a
->vn
| a
->vm
) & 0x10)) {
2729 /* Bad size (including size == 3, which is a different insn group) */
2737 if (!vfp_access_check(s
)) {
2741 scalar
= neon_get_scalar(a
->size
, a
->vm
);
2743 /* Load all inputs before writing any outputs, in case of overlap */
2744 rn
= tcg_temp_new_i32();
2745 read_neon_element32(rn
, a
->vn
, 0, MO_32
);
2746 rn0_64
= tcg_temp_new_i64();
2747 opfn(rn0_64
, rn
, scalar
);
2749 read_neon_element32(rn
, a
->vn
, 1, MO_32
);
2750 rn1_64
= tcg_temp_new_i64();
2751 opfn(rn1_64
, rn
, scalar
);
2752 tcg_temp_free_i32(rn
);
2753 tcg_temp_free_i32(scalar
);
2756 TCGv_i64 t64
= tcg_temp_new_i64();
2757 read_neon_element64(t64
, a
->vd
, 0, MO_64
);
2758 accfn(rn0_64
, t64
, rn0_64
);
2759 read_neon_element64(t64
, a
->vd
, 1, MO_64
);
2760 accfn(rn1_64
, t64
, rn1_64
);
2761 tcg_temp_free_i64(t64
);
2764 write_neon_element64(rn0_64
, a
->vd
, 0, MO_64
);
2765 write_neon_element64(rn1_64
, a
->vd
, 1, MO_64
);
2766 tcg_temp_free_i64(rn0_64
);
2767 tcg_temp_free_i64(rn1_64
);
2771 static bool trans_VMULL_S_2sc(DisasContext
*s
, arg_2scalar
*a
)
2773 static NeonGenTwoOpWidenFn
* const opfn
[] = {
2775 gen_helper_neon_mull_s16
,
2780 return do_2scalar_long(s
, a
, opfn
[a
->size
], NULL
);
2783 static bool trans_VMULL_U_2sc(DisasContext
*s
, arg_2scalar
*a
)
2785 static NeonGenTwoOpWidenFn
* const opfn
[] = {
2787 gen_helper_neon_mull_u16
,
2792 return do_2scalar_long(s
, a
, opfn
[a
->size
], NULL
);
2795 #define DO_VMLAL_2SC(INSN, MULL, ACC) \
2796 static bool trans_##INSN##_2sc(DisasContext *s, arg_2scalar *a) \
2798 static NeonGenTwoOpWidenFn * const opfn[] = { \
2800 gen_helper_neon_##MULL##16, \
2804 static NeonGenTwo64OpFn * const accfn[] = { \
2806 gen_helper_neon_##ACC##l_u32, \
2807 tcg_gen_##ACC##_i64, \
2810 return do_2scalar_long(s, a, opfn[a->size], accfn[a->size]); \
2813 DO_VMLAL_2SC(VMLAL_S
, mull_s
, add
)
2814 DO_VMLAL_2SC(VMLAL_U
, mull_u
, add
)
2815 DO_VMLAL_2SC(VMLSL_S
, mull_s
, sub
)
2816 DO_VMLAL_2SC(VMLSL_U
, mull_u
, sub
)
2818 static bool trans_VQDMULL_2sc(DisasContext
*s
, arg_2scalar
*a
)
2820 static NeonGenTwoOpWidenFn
* const opfn
[] = {
2827 return do_2scalar_long(s
, a
, opfn
[a
->size
], NULL
);
2830 static bool trans_VQDMLAL_2sc(DisasContext
*s
, arg_2scalar
*a
)
2832 static NeonGenTwoOpWidenFn
* const opfn
[] = {
2838 static NeonGenTwo64OpFn
* const accfn
[] = {
2845 return do_2scalar_long(s
, a
, opfn
[a
->size
], accfn
[a
->size
]);
2848 static bool trans_VQDMLSL_2sc(DisasContext
*s
, arg_2scalar
*a
)
2850 static NeonGenTwoOpWidenFn
* const opfn
[] = {
2856 static NeonGenTwo64OpFn
* const accfn
[] = {
2863 return do_2scalar_long(s
, a
, opfn
[a
->size
], accfn
[a
->size
]);
2866 static bool trans_VEXT(DisasContext
*s
, arg_VEXT
*a
)
2868 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
2872 /* UNDEF accesses to D16-D31 if they don't exist. */
2873 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
2874 ((a
->vd
| a
->vn
| a
->vm
) & 0x10)) {
2878 if ((a
->vn
| a
->vm
| a
->vd
) & a
->q
) {
2882 if (a
->imm
> 7 && !a
->q
) {
2886 if (!vfp_access_check(s
)) {
2891 /* Extract 64 bits from <Vm:Vn> */
2892 TCGv_i64 left
, right
, dest
;
2894 left
= tcg_temp_new_i64();
2895 right
= tcg_temp_new_i64();
2896 dest
= tcg_temp_new_i64();
2898 read_neon_element64(right
, a
->vn
, 0, MO_64
);
2899 read_neon_element64(left
, a
->vm
, 0, MO_64
);
2900 tcg_gen_extract2_i64(dest
, right
, left
, a
->imm
* 8);
2901 write_neon_element64(dest
, a
->vd
, 0, MO_64
);
2903 tcg_temp_free_i64(left
);
2904 tcg_temp_free_i64(right
);
2905 tcg_temp_free_i64(dest
);
2907 /* Extract 128 bits from <Vm+1:Vm:Vn+1:Vn> */
2908 TCGv_i64 left
, middle
, right
, destleft
, destright
;
2910 left
= tcg_temp_new_i64();
2911 middle
= tcg_temp_new_i64();
2912 right
= tcg_temp_new_i64();
2913 destleft
= tcg_temp_new_i64();
2914 destright
= tcg_temp_new_i64();
2917 read_neon_element64(right
, a
->vn
, 0, MO_64
);
2918 read_neon_element64(middle
, a
->vn
, 1, MO_64
);
2919 tcg_gen_extract2_i64(destright
, right
, middle
, a
->imm
* 8);
2920 read_neon_element64(left
, a
->vm
, 0, MO_64
);
2921 tcg_gen_extract2_i64(destleft
, middle
, left
, a
->imm
* 8);
2923 read_neon_element64(right
, a
->vn
, 1, MO_64
);
2924 read_neon_element64(middle
, a
->vm
, 0, MO_64
);
2925 tcg_gen_extract2_i64(destright
, right
, middle
, (a
->imm
- 8) * 8);
2926 read_neon_element64(left
, a
->vm
, 1, MO_64
);
2927 tcg_gen_extract2_i64(destleft
, middle
, left
, (a
->imm
- 8) * 8);
2930 write_neon_element64(destright
, a
->vd
, 0, MO_64
);
2931 write_neon_element64(destleft
, a
->vd
, 1, MO_64
);
2933 tcg_temp_free_i64(destright
);
2934 tcg_temp_free_i64(destleft
);
2935 tcg_temp_free_i64(right
);
2936 tcg_temp_free_i64(middle
);
2937 tcg_temp_free_i64(left
);
2942 static bool trans_VTBL(DisasContext
*s
, arg_VTBL
*a
)
2947 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
2951 /* UNDEF accesses to D16-D31 if they don't exist. */
2952 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
2953 ((a
->vd
| a
->vn
| a
->vm
) & 0x10)) {
2957 if ((a
->vn
+ a
->len
+ 1) > 32) {
2959 * This is UNPREDICTABLE; we choose to UNDEF to avoid the
2960 * helper function running off the end of the register file.
2965 if (!vfp_access_check(s
)) {
2969 desc
= tcg_const_i32((a
->vn
<< 2) | a
->len
);
2970 def
= tcg_temp_new_i64();
2972 read_neon_element64(def
, a
->vd
, 0, MO_64
);
2974 tcg_gen_movi_i64(def
, 0);
2976 val
= tcg_temp_new_i64();
2977 read_neon_element64(val
, a
->vm
, 0, MO_64
);
2979 gen_helper_neon_tbl(val
, cpu_env
, desc
, val
, def
);
2980 write_neon_element64(val
, a
->vd
, 0, MO_64
);
2982 tcg_temp_free_i64(def
);
2983 tcg_temp_free_i64(val
);
2984 tcg_temp_free_i32(desc
);
2988 static bool trans_VDUP_scalar(DisasContext
*s
, arg_VDUP_scalar
*a
)
2990 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
2994 /* UNDEF accesses to D16-D31 if they don't exist. */
2995 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
2996 ((a
->vd
| a
->vm
) & 0x10)) {
3004 if (!vfp_access_check(s
)) {
3008 tcg_gen_gvec_dup_mem(a
->size
, neon_full_reg_offset(a
->vd
),
3009 neon_element_offset(a
->vm
, a
->index
, a
->size
),
3010 a
->q
? 16 : 8, a
->q
? 16 : 8);
3014 static bool trans_VREV64(DisasContext
*s
, arg_VREV64
*a
)
3019 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
3023 /* UNDEF accesses to D16-D31 if they don't exist. */
3024 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
3025 ((a
->vd
| a
->vm
) & 0x10)) {
3029 if ((a
->vd
| a
->vm
) & a
->q
) {
3037 if (!vfp_access_check(s
)) {
3041 tmp
[0] = tcg_temp_new_i32();
3042 tmp
[1] = tcg_temp_new_i32();
3044 for (pass
= 0; pass
< (a
->q
? 2 : 1); pass
++) {
3045 for (half
= 0; half
< 2; half
++) {
3046 read_neon_element32(tmp
[half
], a
->vm
, pass
* 2 + half
, MO_32
);
3049 tcg_gen_bswap32_i32(tmp
[half
], tmp
[half
]);
3052 gen_swap_half(tmp
[half
], tmp
[half
]);
3057 g_assert_not_reached();
3060 write_neon_element32(tmp
[1], a
->vd
, pass
* 2, MO_32
);
3061 write_neon_element32(tmp
[0], a
->vd
, pass
* 2 + 1, MO_32
);
3064 tcg_temp_free_i32(tmp
[0]);
3065 tcg_temp_free_i32(tmp
[1]);
3069 static bool do_2misc_pairwise(DisasContext
*s
, arg_2misc
*a
,
3070 NeonGenWidenFn
*widenfn
,
3071 NeonGenTwo64OpFn
*opfn
,
3072 NeonGenTwo64OpFn
*accfn
)
3075 * Pairwise long operations: widen both halves of the pair,
3076 * combine the pairs with the opfn, and then possibly accumulate
3077 * into the destination with the accfn.
3081 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
3085 /* UNDEF accesses to D16-D31 if they don't exist. */
3086 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
3087 ((a
->vd
| a
->vm
) & 0x10)) {
3091 if ((a
->vd
| a
->vm
) & a
->q
) {
3099 if (!vfp_access_check(s
)) {
3103 for (pass
= 0; pass
< a
->q
+ 1; pass
++) {
3105 TCGv_i64 rm0_64
, rm1_64
, rd_64
;
3107 rm0_64
= tcg_temp_new_i64();
3108 rm1_64
= tcg_temp_new_i64();
3109 rd_64
= tcg_temp_new_i64();
3111 tmp
= tcg_temp_new_i32();
3112 read_neon_element32(tmp
, a
->vm
, pass
* 2, MO_32
);
3113 widenfn(rm0_64
, tmp
);
3114 read_neon_element32(tmp
, a
->vm
, pass
* 2 + 1, MO_32
);
3115 widenfn(rm1_64
, tmp
);
3116 tcg_temp_free_i32(tmp
);
3118 opfn(rd_64
, rm0_64
, rm1_64
);
3119 tcg_temp_free_i64(rm0_64
);
3120 tcg_temp_free_i64(rm1_64
);
3123 TCGv_i64 tmp64
= tcg_temp_new_i64();
3124 read_neon_element64(tmp64
, a
->vd
, pass
, MO_64
);
3125 accfn(rd_64
, tmp64
, rd_64
);
3126 tcg_temp_free_i64(tmp64
);
3128 write_neon_element64(rd_64
, a
->vd
, pass
, MO_64
);
3129 tcg_temp_free_i64(rd_64
);
3134 static bool trans_VPADDL_S(DisasContext
*s
, arg_2misc
*a
)
3136 static NeonGenWidenFn
* const widenfn
[] = {
3137 gen_helper_neon_widen_s8
,
3138 gen_helper_neon_widen_s16
,
3139 tcg_gen_ext_i32_i64
,
3142 static NeonGenTwo64OpFn
* const opfn
[] = {
3143 gen_helper_neon_paddl_u16
,
3144 gen_helper_neon_paddl_u32
,
3149 return do_2misc_pairwise(s
, a
, widenfn
[a
->size
], opfn
[a
->size
], NULL
);
3152 static bool trans_VPADDL_U(DisasContext
*s
, arg_2misc
*a
)
3154 static NeonGenWidenFn
* const widenfn
[] = {
3155 gen_helper_neon_widen_u8
,
3156 gen_helper_neon_widen_u16
,
3157 tcg_gen_extu_i32_i64
,
3160 static NeonGenTwo64OpFn
* const opfn
[] = {
3161 gen_helper_neon_paddl_u16
,
3162 gen_helper_neon_paddl_u32
,
3167 return do_2misc_pairwise(s
, a
, widenfn
[a
->size
], opfn
[a
->size
], NULL
);
3170 static bool trans_VPADAL_S(DisasContext
*s
, arg_2misc
*a
)
3172 static NeonGenWidenFn
* const widenfn
[] = {
3173 gen_helper_neon_widen_s8
,
3174 gen_helper_neon_widen_s16
,
3175 tcg_gen_ext_i32_i64
,
3178 static NeonGenTwo64OpFn
* const opfn
[] = {
3179 gen_helper_neon_paddl_u16
,
3180 gen_helper_neon_paddl_u32
,
3184 static NeonGenTwo64OpFn
* const accfn
[] = {
3185 gen_helper_neon_addl_u16
,
3186 gen_helper_neon_addl_u32
,
3191 return do_2misc_pairwise(s
, a
, widenfn
[a
->size
], opfn
[a
->size
],
3195 static bool trans_VPADAL_U(DisasContext
*s
, arg_2misc
*a
)
3197 static NeonGenWidenFn
* const widenfn
[] = {
3198 gen_helper_neon_widen_u8
,
3199 gen_helper_neon_widen_u16
,
3200 tcg_gen_extu_i32_i64
,
3203 static NeonGenTwo64OpFn
* const opfn
[] = {
3204 gen_helper_neon_paddl_u16
,
3205 gen_helper_neon_paddl_u32
,
3209 static NeonGenTwo64OpFn
* const accfn
[] = {
3210 gen_helper_neon_addl_u16
,
3211 gen_helper_neon_addl_u32
,
3216 return do_2misc_pairwise(s
, a
, widenfn
[a
->size
], opfn
[a
->size
],
3220 typedef void ZipFn(TCGv_ptr
, TCGv_ptr
);
3222 static bool do_zip_uzp(DisasContext
*s
, arg_2misc
*a
,
3227 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
3231 /* UNDEF accesses to D16-D31 if they don't exist. */
3232 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
3233 ((a
->vd
| a
->vm
) & 0x10)) {
3237 if ((a
->vd
| a
->vm
) & a
->q
) {
3242 /* Bad size or size/q combination */
3246 if (!vfp_access_check(s
)) {
3250 pd
= vfp_reg_ptr(true, a
->vd
);
3251 pm
= vfp_reg_ptr(true, a
->vm
);
3253 tcg_temp_free_ptr(pd
);
3254 tcg_temp_free_ptr(pm
);
3258 static bool trans_VUZP(DisasContext
*s
, arg_2misc
*a
)
3260 static ZipFn
* const fn
[2][4] = {
3262 gen_helper_neon_unzip8
,
3263 gen_helper_neon_unzip16
,
3267 gen_helper_neon_qunzip8
,
3268 gen_helper_neon_qunzip16
,
3269 gen_helper_neon_qunzip32
,
3273 return do_zip_uzp(s
, a
, fn
[a
->q
][a
->size
]);
3276 static bool trans_VZIP(DisasContext
*s
, arg_2misc
*a
)
3278 static ZipFn
* const fn
[2][4] = {
3280 gen_helper_neon_zip8
,
3281 gen_helper_neon_zip16
,
3285 gen_helper_neon_qzip8
,
3286 gen_helper_neon_qzip16
,
3287 gen_helper_neon_qzip32
,
3291 return do_zip_uzp(s
, a
, fn
[a
->q
][a
->size
]);
3294 static bool do_vmovn(DisasContext
*s
, arg_2misc
*a
,
3295 NeonGenNarrowEnvFn
*narrowfn
)
3300 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
3304 /* UNDEF accesses to D16-D31 if they don't exist. */
3305 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
3306 ((a
->vd
| a
->vm
) & 0x10)) {
3318 if (!vfp_access_check(s
)) {
3322 rm
= tcg_temp_new_i64();
3323 rd0
= tcg_temp_new_i32();
3324 rd1
= tcg_temp_new_i32();
3326 read_neon_element64(rm
, a
->vm
, 0, MO_64
);
3327 narrowfn(rd0
, cpu_env
, rm
);
3328 read_neon_element64(rm
, a
->vm
, 1, MO_64
);
3329 narrowfn(rd1
, cpu_env
, rm
);
3330 write_neon_element32(rd0
, a
->vd
, 0, MO_32
);
3331 write_neon_element32(rd1
, a
->vd
, 1, MO_32
);
3332 tcg_temp_free_i32(rd0
);
3333 tcg_temp_free_i32(rd1
);
3334 tcg_temp_free_i64(rm
);
3338 #define DO_VMOVN(INSN, FUNC) \
3339 static bool trans_##INSN(DisasContext *s, arg_2misc *a) \
3341 static NeonGenNarrowEnvFn * const narrowfn[] = { \
3347 return do_vmovn(s, a, narrowfn[a->size]); \
3350 DO_VMOVN(VMOVN
, gen_neon_narrow_u
)
3351 DO_VMOVN(VQMOVUN
, gen_helper_neon_unarrow_sat
)
3352 DO_VMOVN(VQMOVN_S
, gen_helper_neon_narrow_sat_s
)
3353 DO_VMOVN(VQMOVN_U
, gen_helper_neon_narrow_sat_u
)
3355 static bool trans_VSHLL(DisasContext
*s
, arg_2misc
*a
)
3359 static NeonGenWidenFn
* const widenfns
[] = {
3360 gen_helper_neon_widen_u8
,
3361 gen_helper_neon_widen_u16
,
3362 tcg_gen_extu_i32_i64
,
3365 NeonGenWidenFn
*widenfn
= widenfns
[a
->size
];
3367 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
3371 /* UNDEF accesses to D16-D31 if they don't exist. */
3372 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
3373 ((a
->vd
| a
->vm
) & 0x10)) {
3385 if (!vfp_access_check(s
)) {
3389 rd
= tcg_temp_new_i64();
3390 rm0
= tcg_temp_new_i32();
3391 rm1
= tcg_temp_new_i32();
3393 read_neon_element32(rm0
, a
->vm
, 0, MO_32
);
3394 read_neon_element32(rm1
, a
->vm
, 1, MO_32
);
3397 tcg_gen_shli_i64(rd
, rd
, 8 << a
->size
);
3398 write_neon_element64(rd
, a
->vd
, 0, MO_64
);
3400 tcg_gen_shli_i64(rd
, rd
, 8 << a
->size
);
3401 write_neon_element64(rd
, a
->vd
, 1, MO_64
);
3403 tcg_temp_free_i64(rd
);
3404 tcg_temp_free_i32(rm0
);
3405 tcg_temp_free_i32(rm1
);
3409 static bool trans_VCVT_F16_F32(DisasContext
*s
, arg_2misc
*a
)
3412 TCGv_i32 ahp
, tmp
, tmp2
, tmp3
;
3414 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
) ||
3415 !dc_isar_feature(aa32_fp16_spconv
, s
)) {
3419 /* UNDEF accesses to D16-D31 if they don't exist. */
3420 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
3421 ((a
->vd
| a
->vm
) & 0x10)) {
3425 if ((a
->vm
& 1) || (a
->size
!= 1)) {
3429 if (!vfp_access_check(s
)) {
3433 fpst
= fpstatus_ptr(FPST_STD
);
3434 ahp
= get_ahp_flag();
3435 tmp
= tcg_temp_new_i32();
3436 read_neon_element32(tmp
, a
->vm
, 0, MO_32
);
3437 gen_helper_vfp_fcvt_f32_to_f16(tmp
, tmp
, fpst
, ahp
);
3438 tmp2
= tcg_temp_new_i32();
3439 read_neon_element32(tmp2
, a
->vm
, 1, MO_32
);
3440 gen_helper_vfp_fcvt_f32_to_f16(tmp2
, tmp2
, fpst
, ahp
);
3441 tcg_gen_shli_i32(tmp2
, tmp2
, 16);
3442 tcg_gen_or_i32(tmp2
, tmp2
, tmp
);
3443 read_neon_element32(tmp
, a
->vm
, 2, MO_32
);
3444 gen_helper_vfp_fcvt_f32_to_f16(tmp
, tmp
, fpst
, ahp
);
3445 tmp3
= tcg_temp_new_i32();
3446 read_neon_element32(tmp3
, a
->vm
, 3, MO_32
);
3447 write_neon_element32(tmp2
, a
->vd
, 0, MO_32
);
3448 tcg_temp_free_i32(tmp2
);
3449 gen_helper_vfp_fcvt_f32_to_f16(tmp3
, tmp3
, fpst
, ahp
);
3450 tcg_gen_shli_i32(tmp3
, tmp3
, 16);
3451 tcg_gen_or_i32(tmp3
, tmp3
, tmp
);
3452 write_neon_element32(tmp3
, a
->vd
, 1, MO_32
);
3453 tcg_temp_free_i32(tmp3
);
3454 tcg_temp_free_i32(tmp
);
3455 tcg_temp_free_i32(ahp
);
3456 tcg_temp_free_ptr(fpst
);
3461 static bool trans_VCVT_F32_F16(DisasContext
*s
, arg_2misc
*a
)
3464 TCGv_i32 ahp
, tmp
, tmp2
, tmp3
;
3466 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
) ||
3467 !dc_isar_feature(aa32_fp16_spconv
, s
)) {
3471 /* UNDEF accesses to D16-D31 if they don't exist. */
3472 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
3473 ((a
->vd
| a
->vm
) & 0x10)) {
3477 if ((a
->vd
& 1) || (a
->size
!= 1)) {
3481 if (!vfp_access_check(s
)) {
3485 fpst
= fpstatus_ptr(FPST_STD
);
3486 ahp
= get_ahp_flag();
3487 tmp3
= tcg_temp_new_i32();
3488 tmp2
= tcg_temp_new_i32();
3489 tmp
= tcg_temp_new_i32();
3490 read_neon_element32(tmp
, a
->vm
, 0, MO_32
);
3491 read_neon_element32(tmp2
, a
->vm
, 1, MO_32
);
3492 tcg_gen_ext16u_i32(tmp3
, tmp
);
3493 gen_helper_vfp_fcvt_f16_to_f32(tmp3
, tmp3
, fpst
, ahp
);
3494 write_neon_element32(tmp3
, a
->vd
, 0, MO_32
);
3495 tcg_gen_shri_i32(tmp
, tmp
, 16);
3496 gen_helper_vfp_fcvt_f16_to_f32(tmp
, tmp
, fpst
, ahp
);
3497 write_neon_element32(tmp
, a
->vd
, 1, MO_32
);
3498 tcg_temp_free_i32(tmp
);
3499 tcg_gen_ext16u_i32(tmp3
, tmp2
);
3500 gen_helper_vfp_fcvt_f16_to_f32(tmp3
, tmp3
, fpst
, ahp
);
3501 write_neon_element32(tmp3
, a
->vd
, 2, MO_32
);
3502 tcg_temp_free_i32(tmp3
);
3503 tcg_gen_shri_i32(tmp2
, tmp2
, 16);
3504 gen_helper_vfp_fcvt_f16_to_f32(tmp2
, tmp2
, fpst
, ahp
);
3505 write_neon_element32(tmp2
, a
->vd
, 3, MO_32
);
3506 tcg_temp_free_i32(tmp2
);
3507 tcg_temp_free_i32(ahp
);
3508 tcg_temp_free_ptr(fpst
);
3513 static bool do_2misc_vec(DisasContext
*s
, arg_2misc
*a
, GVecGen2Fn
*fn
)
3515 int vec_size
= a
->q
? 16 : 8;
3516 int rd_ofs
= neon_full_reg_offset(a
->vd
);
3517 int rm_ofs
= neon_full_reg_offset(a
->vm
);
3519 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
3523 /* UNDEF accesses to D16-D31 if they don't exist. */
3524 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
3525 ((a
->vd
| a
->vm
) & 0x10)) {
3533 if ((a
->vd
| a
->vm
) & a
->q
) {
3537 if (!vfp_access_check(s
)) {
3541 fn(a
->size
, rd_ofs
, rm_ofs
, vec_size
, vec_size
);
3546 #define DO_2MISC_VEC(INSN, FN) \
3547 static bool trans_##INSN(DisasContext *s, arg_2misc *a) \
3549 return do_2misc_vec(s, a, FN); \
3552 DO_2MISC_VEC(VNEG
, tcg_gen_gvec_neg
)
3553 DO_2MISC_VEC(VABS
, tcg_gen_gvec_abs
)
3554 DO_2MISC_VEC(VCEQ0
, gen_gvec_ceq0
)
3555 DO_2MISC_VEC(VCGT0
, gen_gvec_cgt0
)
3556 DO_2MISC_VEC(VCLE0
, gen_gvec_cle0
)
3557 DO_2MISC_VEC(VCGE0
, gen_gvec_cge0
)
3558 DO_2MISC_VEC(VCLT0
, gen_gvec_clt0
)
3560 static bool trans_VMVN(DisasContext
*s
, arg_2misc
*a
)
3565 return do_2misc_vec(s
, a
, tcg_gen_gvec_not
);
3568 #define WRAP_2M_3_OOL_FN(WRAPNAME, FUNC, DATA) \
3569 static void WRAPNAME(unsigned vece, uint32_t rd_ofs, \
3570 uint32_t rm_ofs, uint32_t oprsz, \
3573 tcg_gen_gvec_3_ool(rd_ofs, rd_ofs, rm_ofs, oprsz, maxsz, \
3577 #define WRAP_2M_2_OOL_FN(WRAPNAME, FUNC, DATA) \
3578 static void WRAPNAME(unsigned vece, uint32_t rd_ofs, \
3579 uint32_t rm_ofs, uint32_t oprsz, \
3582 tcg_gen_gvec_2_ool(rd_ofs, rm_ofs, oprsz, maxsz, DATA, FUNC); \
3585 WRAP_2M_3_OOL_FN(gen_AESE
, gen_helper_crypto_aese
, 0)
3586 WRAP_2M_3_OOL_FN(gen_AESD
, gen_helper_crypto_aese
, 1)
3587 WRAP_2M_2_OOL_FN(gen_AESMC
, gen_helper_crypto_aesmc
, 0)
3588 WRAP_2M_2_OOL_FN(gen_AESIMC
, gen_helper_crypto_aesmc
, 1)
3589 WRAP_2M_2_OOL_FN(gen_SHA1H
, gen_helper_crypto_sha1h
, 0)
3590 WRAP_2M_2_OOL_FN(gen_SHA1SU1
, gen_helper_crypto_sha1su1
, 0)
3591 WRAP_2M_2_OOL_FN(gen_SHA256SU0
, gen_helper_crypto_sha256su0
, 0)
3593 #define DO_2M_CRYPTO(INSN, FEATURE, SIZE) \
3594 static bool trans_##INSN(DisasContext *s, arg_2misc *a) \
3596 if (!dc_isar_feature(FEATURE, s) || a->size != SIZE) { \
3599 return do_2misc_vec(s, a, gen_##INSN); \
3602 DO_2M_CRYPTO(AESE
, aa32_aes
, 0)
3603 DO_2M_CRYPTO(AESD
, aa32_aes
, 0)
3604 DO_2M_CRYPTO(AESMC
, aa32_aes
, 0)
3605 DO_2M_CRYPTO(AESIMC
, aa32_aes
, 0)
3606 DO_2M_CRYPTO(SHA1H
, aa32_sha1
, 2)
3607 DO_2M_CRYPTO(SHA1SU1
, aa32_sha1
, 2)
3608 DO_2M_CRYPTO(SHA256SU0
, aa32_sha2
, 2)
3610 static bool do_2misc(DisasContext
*s
, arg_2misc
*a
, NeonGenOneOpFn
*fn
)
3615 /* Handle a 2-reg-misc operation by iterating 32 bits at a time */
3616 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
3620 /* UNDEF accesses to D16-D31 if they don't exist. */
3621 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
3622 ((a
->vd
| a
->vm
) & 0x10)) {
3630 if ((a
->vd
| a
->vm
) & a
->q
) {
3634 if (!vfp_access_check(s
)) {
3638 tmp
= tcg_temp_new_i32();
3639 for (pass
= 0; pass
< (a
->q
? 4 : 2); pass
++) {
3640 read_neon_element32(tmp
, a
->vm
, pass
, MO_32
);
3642 write_neon_element32(tmp
, a
->vd
, pass
, MO_32
);
3644 tcg_temp_free_i32(tmp
);
3649 static bool trans_VREV32(DisasContext
*s
, arg_2misc
*a
)
3651 static NeonGenOneOpFn
* const fn
[] = {
3652 tcg_gen_bswap32_i32
,
3657 return do_2misc(s
, a
, fn
[a
->size
]);
3660 static bool trans_VREV16(DisasContext
*s
, arg_2misc
*a
)
3665 return do_2misc(s
, a
, gen_rev16
);
3668 static bool trans_VCLS(DisasContext
*s
, arg_2misc
*a
)
3670 static NeonGenOneOpFn
* const fn
[] = {
3671 gen_helper_neon_cls_s8
,
3672 gen_helper_neon_cls_s16
,
3673 gen_helper_neon_cls_s32
,
3676 return do_2misc(s
, a
, fn
[a
->size
]);
3679 static void do_VCLZ_32(TCGv_i32 rd
, TCGv_i32 rm
)
3681 tcg_gen_clzi_i32(rd
, rm
, 32);
3684 static bool trans_VCLZ(DisasContext
*s
, arg_2misc
*a
)
3686 static NeonGenOneOpFn
* const fn
[] = {
3687 gen_helper_neon_clz_u8
,
3688 gen_helper_neon_clz_u16
,
3692 return do_2misc(s
, a
, fn
[a
->size
]);
3695 static bool trans_VCNT(DisasContext
*s
, arg_2misc
*a
)
3700 return do_2misc(s
, a
, gen_helper_neon_cnt_u8
);
3703 static void gen_VABS_F(unsigned vece
, uint32_t rd_ofs
, uint32_t rm_ofs
,
3704 uint32_t oprsz
, uint32_t maxsz
)
3706 tcg_gen_gvec_andi(vece
, rd_ofs
, rm_ofs
,
3707 vece
== MO_16
? 0x7fff : 0x7fffffff,
3711 static bool trans_VABS_F(DisasContext
*s
, arg_2misc
*a
)
3713 if (a
->size
== MO_16
) {
3714 if (!dc_isar_feature(aa32_fp16_arith
, s
)) {
3717 } else if (a
->size
!= MO_32
) {
3720 return do_2misc_vec(s
, a
, gen_VABS_F
);
3723 static void gen_VNEG_F(unsigned vece
, uint32_t rd_ofs
, uint32_t rm_ofs
,
3724 uint32_t oprsz
, uint32_t maxsz
)
3726 tcg_gen_gvec_xori(vece
, rd_ofs
, rm_ofs
,
3727 vece
== MO_16
? 0x8000 : 0x80000000,
3731 static bool trans_VNEG_F(DisasContext
*s
, arg_2misc
*a
)
3733 if (a
->size
== MO_16
) {
3734 if (!dc_isar_feature(aa32_fp16_arith
, s
)) {
3737 } else if (a
->size
!= MO_32
) {
3740 return do_2misc_vec(s
, a
, gen_VNEG_F
);
3743 static bool trans_VRECPE(DisasContext
*s
, arg_2misc
*a
)
3748 return do_2misc(s
, a
, gen_helper_recpe_u32
);
3751 static bool trans_VRSQRTE(DisasContext
*s
, arg_2misc
*a
)
3756 return do_2misc(s
, a
, gen_helper_rsqrte_u32
);
3759 #define WRAP_1OP_ENV_FN(WRAPNAME, FUNC) \
3760 static void WRAPNAME(TCGv_i32 d, TCGv_i32 m) \
3762 FUNC(d, cpu_env, m); \
3765 WRAP_1OP_ENV_FN(gen_VQABS_s8
, gen_helper_neon_qabs_s8
)
3766 WRAP_1OP_ENV_FN(gen_VQABS_s16
, gen_helper_neon_qabs_s16
)
3767 WRAP_1OP_ENV_FN(gen_VQABS_s32
, gen_helper_neon_qabs_s32
)
3768 WRAP_1OP_ENV_FN(gen_VQNEG_s8
, gen_helper_neon_qneg_s8
)
3769 WRAP_1OP_ENV_FN(gen_VQNEG_s16
, gen_helper_neon_qneg_s16
)
3770 WRAP_1OP_ENV_FN(gen_VQNEG_s32
, gen_helper_neon_qneg_s32
)
3772 static bool trans_VQABS(DisasContext
*s
, arg_2misc
*a
)
3774 static NeonGenOneOpFn
* const fn
[] = {
3780 return do_2misc(s
, a
, fn
[a
->size
]);
3783 static bool trans_VQNEG(DisasContext
*s
, arg_2misc
*a
)
3785 static NeonGenOneOpFn
* const fn
[] = {
3791 return do_2misc(s
, a
, fn
[a
->size
]);
3794 #define DO_2MISC_FP_VEC(INSN, HFUNC, SFUNC) \
3795 static void gen_##INSN(unsigned vece, uint32_t rd_ofs, \
3797 uint32_t oprsz, uint32_t maxsz) \
3799 static gen_helper_gvec_2_ptr * const fns[4] = { \
3800 NULL, HFUNC, SFUNC, NULL, \
3803 fpst = fpstatus_ptr(vece == MO_16 ? FPST_STD_F16 : FPST_STD); \
3804 tcg_gen_gvec_2_ptr(rd_ofs, rm_ofs, fpst, oprsz, maxsz, 0, \
3806 tcg_temp_free_ptr(fpst); \
3808 static bool trans_##INSN(DisasContext *s, arg_2misc *a) \
3810 if (a->size == MO_16) { \
3811 if (!dc_isar_feature(aa32_fp16_arith, s)) { \
3814 } else if (a->size != MO_32) { \
3817 return do_2misc_vec(s, a, gen_##INSN); \
3820 DO_2MISC_FP_VEC(VRECPE_F
, gen_helper_gvec_frecpe_h
, gen_helper_gvec_frecpe_s
)
3821 DO_2MISC_FP_VEC(VRSQRTE_F
, gen_helper_gvec_frsqrte_h
, gen_helper_gvec_frsqrte_s
)
3822 DO_2MISC_FP_VEC(VCGT0_F
, gen_helper_gvec_fcgt0_h
, gen_helper_gvec_fcgt0_s
)
3823 DO_2MISC_FP_VEC(VCGE0_F
, gen_helper_gvec_fcge0_h
, gen_helper_gvec_fcge0_s
)
3824 DO_2MISC_FP_VEC(VCEQ0_F
, gen_helper_gvec_fceq0_h
, gen_helper_gvec_fceq0_s
)
3825 DO_2MISC_FP_VEC(VCLT0_F
, gen_helper_gvec_fclt0_h
, gen_helper_gvec_fclt0_s
)
3826 DO_2MISC_FP_VEC(VCLE0_F
, gen_helper_gvec_fcle0_h
, gen_helper_gvec_fcle0_s
)
3827 DO_2MISC_FP_VEC(VCVT_FS
, gen_helper_gvec_sstoh
, gen_helper_gvec_sitos
)
3828 DO_2MISC_FP_VEC(VCVT_FU
, gen_helper_gvec_ustoh
, gen_helper_gvec_uitos
)
3829 DO_2MISC_FP_VEC(VCVT_SF
, gen_helper_gvec_tosszh
, gen_helper_gvec_tosizs
)
3830 DO_2MISC_FP_VEC(VCVT_UF
, gen_helper_gvec_touszh
, gen_helper_gvec_touizs
)
3832 DO_2MISC_FP_VEC(VRINTX_impl
, gen_helper_gvec_vrintx_h
, gen_helper_gvec_vrintx_s
)
3834 static bool trans_VRINTX(DisasContext
*s
, arg_2misc
*a
)
3836 if (!arm_dc_feature(s
, ARM_FEATURE_V8
)) {
3839 return trans_VRINTX_impl(s
, a
);
3842 #define DO_VEC_RMODE(INSN, RMODE, OP) \
3843 static void gen_##INSN(unsigned vece, uint32_t rd_ofs, \
3845 uint32_t oprsz, uint32_t maxsz) \
3847 static gen_helper_gvec_2_ptr * const fns[4] = { \
3849 gen_helper_gvec_##OP##h, \
3850 gen_helper_gvec_##OP##s, \
3854 fpst = fpstatus_ptr(vece == 1 ? FPST_STD_F16 : FPST_STD); \
3855 tcg_gen_gvec_2_ptr(rd_ofs, rm_ofs, fpst, oprsz, maxsz, \
3856 arm_rmode_to_sf(RMODE), fns[vece]); \
3857 tcg_temp_free_ptr(fpst); \
3859 static bool trans_##INSN(DisasContext *s, arg_2misc *a) \
3861 if (!arm_dc_feature(s, ARM_FEATURE_V8)) { \
3864 if (a->size == MO_16) { \
3865 if (!dc_isar_feature(aa32_fp16_arith, s)) { \
3868 } else if (a->size != MO_32) { \
3871 return do_2misc_vec(s, a, gen_##INSN); \
3874 DO_VEC_RMODE(VCVTAU
, FPROUNDING_TIEAWAY
, vcvt_rm_u
)
3875 DO_VEC_RMODE(VCVTAS
, FPROUNDING_TIEAWAY
, vcvt_rm_s
)
3876 DO_VEC_RMODE(VCVTNU
, FPROUNDING_TIEEVEN
, vcvt_rm_u
)
3877 DO_VEC_RMODE(VCVTNS
, FPROUNDING_TIEEVEN
, vcvt_rm_s
)
3878 DO_VEC_RMODE(VCVTPU
, FPROUNDING_POSINF
, vcvt_rm_u
)
3879 DO_VEC_RMODE(VCVTPS
, FPROUNDING_POSINF
, vcvt_rm_s
)
3880 DO_VEC_RMODE(VCVTMU
, FPROUNDING_NEGINF
, vcvt_rm_u
)
3881 DO_VEC_RMODE(VCVTMS
, FPROUNDING_NEGINF
, vcvt_rm_s
)
3883 DO_VEC_RMODE(VRINTN
, FPROUNDING_TIEEVEN
, vrint_rm_
)
3884 DO_VEC_RMODE(VRINTA
, FPROUNDING_TIEAWAY
, vrint_rm_
)
3885 DO_VEC_RMODE(VRINTZ
, FPROUNDING_ZERO
, vrint_rm_
)
3886 DO_VEC_RMODE(VRINTM
, FPROUNDING_NEGINF
, vrint_rm_
)
3887 DO_VEC_RMODE(VRINTP
, FPROUNDING_POSINF
, vrint_rm_
)
3889 static bool trans_VSWP(DisasContext
*s
, arg_2misc
*a
)
3894 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
3898 /* UNDEF accesses to D16-D31 if they don't exist. */
3899 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
3900 ((a
->vd
| a
->vm
) & 0x10)) {
3908 if ((a
->vd
| a
->vm
) & a
->q
) {
3912 if (!vfp_access_check(s
)) {
3916 rm
= tcg_temp_new_i64();
3917 rd
= tcg_temp_new_i64();
3918 for (pass
= 0; pass
< (a
->q
? 2 : 1); pass
++) {
3919 read_neon_element64(rm
, a
->vm
, pass
, MO_64
);
3920 read_neon_element64(rd
, a
->vd
, pass
, MO_64
);
3921 write_neon_element64(rm
, a
->vd
, pass
, MO_64
);
3922 write_neon_element64(rd
, a
->vm
, pass
, MO_64
);
3924 tcg_temp_free_i64(rm
);
3925 tcg_temp_free_i64(rd
);
3929 static void gen_neon_trn_u8(TCGv_i32 t0
, TCGv_i32 t1
)
3933 rd
= tcg_temp_new_i32();
3934 tmp
= tcg_temp_new_i32();
3936 tcg_gen_shli_i32(rd
, t0
, 8);
3937 tcg_gen_andi_i32(rd
, rd
, 0xff00ff00);
3938 tcg_gen_andi_i32(tmp
, t1
, 0x00ff00ff);
3939 tcg_gen_or_i32(rd
, rd
, tmp
);
3941 tcg_gen_shri_i32(t1
, t1
, 8);
3942 tcg_gen_andi_i32(t1
, t1
, 0x00ff00ff);
3943 tcg_gen_andi_i32(tmp
, t0
, 0xff00ff00);
3944 tcg_gen_or_i32(t1
, t1
, tmp
);
3945 tcg_gen_mov_i32(t0
, rd
);
3947 tcg_temp_free_i32(tmp
);
3948 tcg_temp_free_i32(rd
);
3951 static void gen_neon_trn_u16(TCGv_i32 t0
, TCGv_i32 t1
)
3955 rd
= tcg_temp_new_i32();
3956 tmp
= tcg_temp_new_i32();
3958 tcg_gen_shli_i32(rd
, t0
, 16);
3959 tcg_gen_andi_i32(tmp
, t1
, 0xffff);
3960 tcg_gen_or_i32(rd
, rd
, tmp
);
3961 tcg_gen_shri_i32(t1
, t1
, 16);
3962 tcg_gen_andi_i32(tmp
, t0
, 0xffff0000);
3963 tcg_gen_or_i32(t1
, t1
, tmp
);
3964 tcg_gen_mov_i32(t0
, rd
);
3966 tcg_temp_free_i32(tmp
);
3967 tcg_temp_free_i32(rd
);
3970 static bool trans_VTRN(DisasContext
*s
, arg_2misc
*a
)
3975 if (!arm_dc_feature(s
, ARM_FEATURE_NEON
)) {
3979 /* UNDEF accesses to D16-D31 if they don't exist. */
3980 if (!dc_isar_feature(aa32_simd_r32
, s
) &&
3981 ((a
->vd
| a
->vm
) & 0x10)) {
3985 if ((a
->vd
| a
->vm
) & a
->q
) {
3993 if (!vfp_access_check(s
)) {
3997 tmp
= tcg_temp_new_i32();
3998 tmp2
= tcg_temp_new_i32();
3999 if (a
->size
== MO_32
) {
4000 for (pass
= 0; pass
< (a
->q
? 4 : 2); pass
+= 2) {
4001 read_neon_element32(tmp
, a
->vm
, pass
, MO_32
);
4002 read_neon_element32(tmp2
, a
->vd
, pass
+ 1, MO_32
);
4003 write_neon_element32(tmp2
, a
->vm
, pass
, MO_32
);
4004 write_neon_element32(tmp
, a
->vd
, pass
+ 1, MO_32
);
4007 for (pass
= 0; pass
< (a
->q
? 4 : 2); pass
++) {
4008 read_neon_element32(tmp
, a
->vm
, pass
, MO_32
);
4009 read_neon_element32(tmp2
, a
->vd
, pass
, MO_32
);
4010 if (a
->size
== MO_8
) {
4011 gen_neon_trn_u8(tmp
, tmp2
);
4013 gen_neon_trn_u16(tmp
, tmp2
);
4015 write_neon_element32(tmp2
, a
->vm
, pass
, MO_32
);
4016 write_neon_element32(tmp
, a
->vd
, pass
, MO_32
);
4019 tcg_temp_free_i32(tmp
);
4020 tcg_temp_free_i32(tmp2
);