@@ -718,6 +718,26 @@ DEF_HELPER_FLAGS_3(gvec_usra_h, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
DEF_HELPER_FLAGS_3(gvec_usra_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
DEF_HELPER_FLAGS_3(gvec_usra_d, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_srshr_b, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_srshr_h, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_srshr_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_srshr_d, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+
+DEF_HELPER_FLAGS_3(gvec_urshr_b, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_urshr_h, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_urshr_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_urshr_d, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+
+DEF_HELPER_FLAGS_3(gvec_srsra_b, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_srsra_h, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_srsra_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_srsra_d, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+
+DEF_HELPER_FLAGS_3(gvec_ursra_b, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_ursra_h, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_ursra_s, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+DEF_HELPER_FLAGS_3(gvec_ursra_d, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
+
#ifdef TARGET_AARCH64
#include "helper-a64.h"
#include "helper-sve.h"
@@ -129,4 +129,13 @@ typedef void GVecGen3Fn(unsigned, uint32_t, uint32_t,
typedef void GVecGen4Fn(unsigned, uint32_t, uint32_t, uint32_t,
uint32_t, uint32_t, uint32_t);
+void arm_gen_gvec_srshr(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ int64_t shift, uint32_t opr_sz, uint32_t max_sz);
+void arm_gen_gvec_urshr(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ int64_t shift, uint32_t opr_sz, uint32_t max_sz);
+void arm_gen_gvec_srsra(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ int64_t shift, uint32_t opr_sz, uint32_t max_sz);
+void arm_gen_gvec_ursra(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ int64_t shift, uint32_t opr_sz, uint32_t max_sz);
+
#endif /* TARGET_ARM_TRANSLATE_A64_H */
@@ -8561,6 +8561,453 @@ static void handle_shri_with_rndacc(TCGv_i64 tcg_res, TCGv_i64 tcg_src,
}
}
+static void gen_srshr8_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
+{
+ TCGv_i64 t = tcg_temp_new_i64();
+ TCGv_i64 ones = tcg_const_i64(dup_const(MO_8, 1));
+
+ /* Shift one less than the requested amount. */
+ if (shift > 1) {
+ tcg_gen_vec_sar8i_i64(a, a, shift - 1);
+ }
+
+ /* The low bit is the rounding bit. Mask it off. */
+ tcg_gen_and_i64(t, a, ones);
+
+ /* Finish the shift. */
+ tcg_gen_vec_sar8i_i64(d, a, 1);
+
+ /* Round. */
+ tcg_gen_vec_add8_i64(d, d, t);
+
+ tcg_temp_free_i64(t);
+ tcg_temp_free_i64(ones);
+}
+
+static void gen_srshr16_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
+{
+ TCGv_i64 t = tcg_temp_new_i64();
+ TCGv_i64 ones = tcg_const_i64(dup_const(MO_16, 1));
+
+ if (shift > 1) {
+ tcg_gen_vec_sar16i_i64(a, a, shift - 1);
+ }
+ tcg_gen_and_i64(t, a, ones);
+ tcg_gen_vec_sar16i_i64(d, a, 1);
+ tcg_gen_vec_add16_i64(d, d, t);
+
+ tcg_temp_free_i64(t);
+ tcg_temp_free_i64(ones);
+}
+
+static void gen_srshr32_i32(TCGv_i32 d, TCGv_i32 a, int32_t shift)
+{
+ TCGv_i32 t = tcg_temp_new_i32();
+
+ tcg_gen_sari_i32(a, a, shift - 1);
+ tcg_gen_andi_i32(t, a, 1);
+ tcg_gen_sari_i32(d, a, 1);
+ tcg_gen_add_i32(d, d, t);
+
+ tcg_temp_free_i32(t);
+}
+
+static void gen_srshr64_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
+{
+ TCGv_i64 t = tcg_temp_new_i64();
+
+ tcg_gen_sari_i64(a, a, shift - 1);
+ tcg_gen_andi_i64(t, a, 1);
+ tcg_gen_sari_i64(d, a, 1);
+ tcg_gen_add_i64(d, d, t);
+
+ tcg_temp_free_i64(t);
+}
+
+static void gen_srshr_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t shift)
+{
+ TCGv_vec t = tcg_temp_new_vec_matching(d);
+ TCGv_vec ones = tcg_temp_new_vec_matching(d);
+
+ tcg_gen_sari_vec(vece, a, a, shift - 1);
+ tcg_gen_dupi_vec(vece, ones, 1);
+ tcg_gen_and_vec(vece, t, a, ones);
+ tcg_gen_sari_vec(vece, d, a, 1);
+ tcg_gen_add_vec(vece, d, d, t);
+
+ tcg_temp_free_vec(t);
+ tcg_temp_free_vec(ones);
+}
+
+void arm_gen_gvec_srshr(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ int64_t shift, uint32_t opr_sz, uint32_t max_sz)
+{
+ static const TCGOpcode vecop_list[] = {
+ INDEX_op_sari_vec, INDEX_op_add_vec, 0
+ };
+ static const GVecGen2i ops[4] = {
+ { .fni8 = gen_srshr8_i64,
+ .fniv = gen_srshr_vec,
+ .fno = gen_helper_gvec_srshr_b,
+ .opt_opc = vecop_list,
+ .vece = MO_8 },
+ { .fni8 = gen_srshr16_i64,
+ .fniv = gen_srshr_vec,
+ .fno = gen_helper_gvec_srshr_h,
+ .opt_opc = vecop_list,
+ .vece = MO_16 },
+ { .fni4 = gen_srshr32_i32,
+ .fniv = gen_srshr_vec,
+ .fno = gen_helper_gvec_srshr_s,
+ .opt_opc = vecop_list,
+ .vece = MO_32 },
+ { .fni8 = gen_srshr64_i64,
+ .fniv = gen_srshr_vec,
+ .fno = gen_helper_gvec_srshr_d,
+ .prefer_i64 = TCG_TARGET_REG_BITS == 64,
+ .opt_opc = vecop_list,
+ .vece = MO_64 },
+ };
+
+ /* tszimm encoding produces immediates in the range [1..esize] */
+ tcg_debug_assert(shift > 0);
+ tcg_debug_assert(shift <= (8 << vece));
+
+ if (shift == (8 << vece)) {
+ /*
+ * Shifts larger than the element size are architecturally valid.
+ * Signed results in all sign bits. With rounding, this produces
+ * (-1 + 1) >> 1 == 0, or (0 + 1) >> 1 == 0.
+ * I.e. always zero.
+ */
+ tcg_gen_gvec_dup8i(rd_ofs, opr_sz, max_sz, 0);
+ } else {
+ tcg_gen_gvec_2i(rd_ofs, rm_ofs, opr_sz, max_sz, shift, &ops[vece]);
+ }
+}
+
+static void gen_srsra8_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
+{
+ TCGv_i64 t = tcg_temp_new_i64();
+ gen_srshr8_i64(t, a, shift);
+ tcg_gen_vec_add8_i64(d, d, t);
+ tcg_temp_free_i64(t);
+}
+
+static void gen_srsra16_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
+{
+ TCGv_i64 t = tcg_temp_new_i64();
+ gen_srshr16_i64(t, a, shift);
+ tcg_gen_vec_add16_i64(d, d, t);
+ tcg_temp_free_i64(t);
+}
+
+static void gen_srsra32_i32(TCGv_i32 d, TCGv_i32 a, int32_t shift)
+{
+ TCGv_i32 t = tcg_temp_new_i32();
+ gen_srshr32_i32(t, a, shift);
+ tcg_gen_add_i32(d, d, t);
+ tcg_temp_free_i32(t);
+}
+
+static void gen_srsra64_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
+{
+ TCGv_i64 t = tcg_temp_new_i64();
+ gen_srshr64_i64(t, a, shift);
+ tcg_gen_add_i64(d, d, t);
+ tcg_temp_free_i64(t);
+}
+
+static void gen_srsra_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t shift)
+{
+ TCGv_vec t = tcg_temp_new_vec_matching(d);
+ gen_srshr_vec(vece, t, a, shift);
+ tcg_gen_add_vec(vece, d, d, t);
+ tcg_temp_free_vec(t);
+}
+
+void arm_gen_gvec_srsra(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ int64_t shift, uint32_t opr_sz, uint32_t max_sz)
+{
+ static const TCGOpcode vecop_list[] = {
+ INDEX_op_sari_vec, INDEX_op_add_vec, 0
+ };
+ static const GVecGen2i ops[4] = {
+ { .fni8 = gen_srsra8_i64,
+ .fniv = gen_srsra_vec,
+ .fno = gen_helper_gvec_srsra_b,
+ .opt_opc = vecop_list,
+ .load_dest = true,
+ .vece = MO_8 },
+ { .fni8 = gen_srsra16_i64,
+ .fniv = gen_srsra_vec,
+ .fno = gen_helper_gvec_srsra_h,
+ .opt_opc = vecop_list,
+ .load_dest = true,
+ .vece = MO_16 },
+ { .fni4 = gen_srsra32_i32,
+ .fniv = gen_srsra_vec,
+ .fno = gen_helper_gvec_srsra_s,
+ .opt_opc = vecop_list,
+ .load_dest = true,
+ .vece = MO_32 },
+ { .fni8 = gen_srsra64_i64,
+ .fniv = gen_srsra_vec,
+ .fno = gen_helper_gvec_srsra_d,
+ .prefer_i64 = TCG_TARGET_REG_BITS == 64,
+ .opt_opc = vecop_list,
+ .load_dest = true,
+ .vece = MO_64 },
+ };
+
+ /* tszimm encoding produces immediates in the range [1..esize] */
+ tcg_debug_assert(shift > 0);
+ tcg_debug_assert(shift <= (8 << vece));
+
+ /*
+ * Shifts larger than the element size are architecturally valid.
+ * Signed results in all sign bits. With rounding, this produces
+ * (-1 + 1) >> 1 == 0, or (0 + 1) >> 1 == 0.
+ * I.e. always zero. With accumulation, this leaves D unchanged.
+ */
+ if (shift != (8 << vece)) {
+ tcg_gen_gvec_2i(rd_ofs, rm_ofs, opr_sz, max_sz, shift, &ops[vece]);
+ } else {
+ /* Nop, but we do need to clear the tail. */
+ tcg_gen_gvec_mov(vece, rd_ofs, rd_ofs, opr_sz, max_sz);
+ }
+}
+
+static void gen_urshr8_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
+{
+ TCGv_i64 t = tcg_temp_new_i64();
+ TCGv_i64 ones = tcg_const_i64(dup_const(MO_8, 1));
+
+ /* Shift one less than the requested amount. */
+ if (shift > 1) {
+ tcg_gen_vec_shr8i_i64(a, a, shift - 1);
+ }
+
+ /* The low bit is the rounding bit. Mask it off. */
+ tcg_gen_and_i64(t, a, ones);
+
+ /* Finish the shift. */
+ tcg_gen_vec_shr8i_i64(d, a, 1);
+
+ /* Round. */
+ tcg_gen_vec_add8_i64(d, d, t);
+
+ tcg_temp_free_i64(t);
+ tcg_temp_free_i64(ones);
+}
+
+static void gen_urshr16_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
+{
+ TCGv_i64 t = tcg_temp_new_i64();
+ TCGv_i64 ones = tcg_const_i64(dup_const(MO_16, 1));
+
+ if (shift > 1) {
+ tcg_gen_vec_shr16i_i64(a, a, shift - 1);
+ }
+ tcg_gen_and_i64(t, a, ones);
+ tcg_gen_vec_shr16i_i64(d, a, 1);
+ tcg_gen_vec_add16_i64(d, d, t);
+
+ tcg_temp_free_i64(t);
+ tcg_temp_free_i64(ones);
+}
+
+static void gen_urshr32_i32(TCGv_i32 d, TCGv_i32 a, int32_t shift)
+{
+ TCGv_i32 t = tcg_temp_new_i32();
+
+ tcg_gen_shri_i32(a, a, shift - 1);
+ tcg_gen_andi_i32(t, a, 1);
+ tcg_gen_shri_i32(d, a, 1);
+ tcg_gen_add_i32(d, d, t);
+
+ tcg_temp_free_i32(t);
+}
+
+static void gen_urshr64_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
+{
+ TCGv_i64 t = tcg_temp_new_i64();
+
+ tcg_gen_shri_i64(a, a, shift - 1);
+ tcg_gen_andi_i64(t, a, 1);
+ tcg_gen_shri_i64(d, a, 1);
+ tcg_gen_add_i64(d, d, t);
+
+ tcg_temp_free_i64(t);
+}
+
+static void gen_urshr_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t shift)
+{
+ TCGv_vec t = tcg_temp_new_vec_matching(d);
+ TCGv_vec ones = tcg_temp_new_vec_matching(d);
+
+ tcg_gen_shri_vec(vece, a, a, shift - 1);
+ tcg_gen_dupi_vec(vece, ones, 1);
+ tcg_gen_and_vec(vece, t, a, ones);
+ tcg_gen_shri_vec(vece, d, a, 1);
+ tcg_gen_add_vec(vece, d, d, t);
+
+ tcg_temp_free_vec(t);
+ tcg_temp_free_vec(ones);
+}
+
+void arm_gen_gvec_urshr(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ int64_t shift, uint32_t opr_sz, uint32_t max_sz)
+{
+ static const TCGOpcode vecop_list[] = {
+ INDEX_op_shri_vec, INDEX_op_add_vec, 0
+ };
+ static const GVecGen2i ops[4] = {
+ { .fni8 = gen_urshr8_i64,
+ .fniv = gen_urshr_vec,
+ .fno = gen_helper_gvec_urshr_b,
+ .opt_opc = vecop_list,
+ .vece = MO_8 },
+ { .fni8 = gen_urshr16_i64,
+ .fniv = gen_urshr_vec,
+ .fno = gen_helper_gvec_urshr_h,
+ .opt_opc = vecop_list,
+ .vece = MO_16 },
+ { .fni4 = gen_urshr32_i32,
+ .fniv = gen_urshr_vec,
+ .fno = gen_helper_gvec_urshr_s,
+ .opt_opc = vecop_list,
+ .vece = MO_32 },
+ { .fni8 = gen_urshr64_i64,
+ .fniv = gen_urshr_vec,
+ .fno = gen_helper_gvec_urshr_d,
+ .prefer_i64 = TCG_TARGET_REG_BITS == 64,
+ .opt_opc = vecop_list,
+ .vece = MO_64 },
+ };
+
+ /* tszimm encoding produces immediates in the range [1..esize] */
+ tcg_debug_assert(shift > 0);
+ tcg_debug_assert(shift <= (8 << vece));
+
+ if (shift == (8 << vece)) {
+ /*
+ * Shifts larger than the element size are architecturally valid.
+ * Unsigned results in zero. With rounding, this produces a
+ * copy of the most significant bit.
+ */
+ tcg_gen_gvec_shri(vece, rd_ofs, rm_ofs, shift - 1, opr_sz, max_sz);
+ } else {
+ tcg_gen_gvec_2i(rd_ofs, rm_ofs, opr_sz, max_sz, shift, &ops[vece]);
+ }
+}
+
+static void gen_ursra8_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
+{
+ TCGv_i64 t = tcg_temp_new_i64();
+
+ if (shift == 8) {
+ tcg_gen_vec_shr8i_i64(t, a, 7);
+ } else {
+ gen_urshr8_i64(t, a, shift);
+ }
+ tcg_gen_vec_add8_i64(d, d, t);
+ tcg_temp_free_i64(t);
+}
+
+static void gen_ursra16_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
+{
+ TCGv_i64 t = tcg_temp_new_i64();
+
+ if (shift == 16) {
+ tcg_gen_vec_shr16i_i64(t, a, 15);
+ } else {
+ gen_urshr16_i64(t, a, shift);
+ }
+ tcg_gen_vec_add16_i64(d, d, t);
+ tcg_temp_free_i64(t);
+}
+
+static void gen_ursra32_i32(TCGv_i32 d, TCGv_i32 a, int32_t shift)
+{
+ TCGv_i32 t = tcg_temp_new_i32();
+
+ if (shift == 32) {
+ tcg_gen_shri_i32(t, a, 31);
+ } else {
+ gen_urshr32_i32(t, a, shift);
+ }
+ tcg_gen_add_i32(d, d, t);
+ tcg_temp_free_i32(t);
+}
+
+static void gen_ursra64_i64(TCGv_i64 d, TCGv_i64 a, int64_t shift)
+{
+ TCGv_i64 t = tcg_temp_new_i64();
+
+ if (shift == 64) {
+ tcg_gen_shri_i64(t, a, 63);
+ } else {
+ gen_urshr64_i64(t, a, shift);
+ }
+ tcg_gen_add_i64(d, d, t);
+ tcg_temp_free_i64(t);
+}
+
+static void gen_ursra_vec(unsigned vece, TCGv_vec d, TCGv_vec a, int64_t shift)
+{
+ TCGv_vec t = tcg_temp_new_vec_matching(d);
+
+ if (shift == (8 << vece)) {
+ tcg_gen_shri_vec(vece, t, a, (8 << vece) - 1);
+ } else {
+ gen_urshr_vec(vece, t, a, shift);
+ }
+ tcg_gen_add_vec(vece, d, d, t);
+ tcg_temp_free_vec(t);
+}
+
+void arm_gen_gvec_ursra(unsigned vece, uint32_t rd_ofs, uint32_t rm_ofs,
+ int64_t shift, uint32_t opr_sz, uint32_t max_sz)
+{
+ static const TCGOpcode vecop_list[] = {
+ INDEX_op_shri_vec, INDEX_op_add_vec, 0
+ };
+ static const GVecGen2i ops[4] = {
+ { .fni8 = gen_ursra8_i64,
+ .fniv = gen_ursra_vec,
+ .fno = gen_helper_gvec_ursra_b,
+ .opt_opc = vecop_list,
+ .load_dest = true,
+ .vece = MO_8 },
+ { .fni8 = gen_ursra16_i64,
+ .fniv = gen_ursra_vec,
+ .fno = gen_helper_gvec_ursra_h,
+ .opt_opc = vecop_list,
+ .load_dest = true,
+ .vece = MO_16 },
+ { .fni4 = gen_ursra32_i32,
+ .fniv = gen_ursra_vec,
+ .fno = gen_helper_gvec_ursra_s,
+ .opt_opc = vecop_list,
+ .load_dest = true,
+ .vece = MO_32 },
+ { .fni8 = gen_ursra64_i64,
+ .fniv = gen_ursra_vec,
+ .fno = gen_helper_gvec_ursra_d,
+ .prefer_i64 = TCG_TARGET_REG_BITS == 64,
+ .opt_opc = vecop_list,
+ .load_dest = true,
+ .vece = MO_64 },
+ };
+
+ /* tszimm encoding produces immediates in the range [1..esize] */
+ tcg_debug_assert(shift > 0);
+ tcg_debug_assert(shift <= (8 << vece));
+
+ tcg_gen_gvec_2i(rd_ofs, rm_ofs, opr_sz, max_sz, shift, &ops[vece]);
+}
+
/* SSHR[RA]/USHR[RA] - Scalar shift right (optional rounding/accumulate) */
static void handle_scalar_simd_shri(DisasContext *s,
bool is_u, int immh, int immb,
@@ -10712,10 +11159,15 @@ static void handle_vec_simd_shri(DisasContext *s, bool is_q, bool is_u,
return;
case 0x04: /* SRSHR / URSHR (rounding) */
- break;
+ gen_gvec_fn2i(s, is_q, rd, rn, shift,
+ is_u ? arm_gen_gvec_urshr : arm_gen_gvec_srshr, size);
+ return;
+
case 0x06: /* SRSRA / URSRA (accum + rounding) */
- accumulate = true;
- break;
+ gen_gvec_fn2i(s, is_q, rd, rn, shift,
+ is_u ? arm_gen_gvec_ursra : arm_gen_gvec_srsra, size);
+ return;
+
default:
g_assert_not_reached();
}
@@ -924,6 +924,56 @@ DO_SRA(gvec_usra_d, uint64_t)
#undef DO_SRA
+#define DO_RSHR(NAME, TYPE) \
+void HELPER(NAME)(void *vd, void *vn, uint32_t desc) \
+{ \
+ intptr_t i, oprsz = simd_oprsz(desc); \
+ int shift = simd_data(desc); \
+ TYPE *d = vd, *n = vn; \
+ for (i = 0; i < oprsz / sizeof(TYPE); i++) { \
+ TYPE tmp = n[i] >> (shift - 1); \
+ d[i] = (tmp >> 1) + (tmp & 1); \
+ } \
+ clear_tail(d, oprsz, simd_maxsz(desc)); \
+}
+
+DO_RSHR(gvec_srshr_b, int8_t)
+DO_RSHR(gvec_srshr_h, int16_t)
+DO_RSHR(gvec_srshr_s, int32_t)
+DO_RSHR(gvec_srshr_d, int64_t)
+
+DO_RSHR(gvec_urshr_b, uint8_t)
+DO_RSHR(gvec_urshr_h, uint16_t)
+DO_RSHR(gvec_urshr_s, uint32_t)
+DO_RSHR(gvec_urshr_d, uint64_t)
+
+#undef DO_RSHR
+
+#define DO_RSRA(NAME, TYPE) \
+void HELPER(NAME)(void *vd, void *vn, uint32_t desc) \
+{ \
+ intptr_t i, oprsz = simd_oprsz(desc); \
+ int shift = simd_data(desc); \
+ TYPE *d = vd, *n = vn; \
+ for (i = 0; i < oprsz / sizeof(TYPE); i++) { \
+ TYPE tmp = n[i] >> (shift - 1); \
+ d[i] += (tmp >> 1) + (tmp & 1); \
+ } \
+ clear_tail(d, oprsz, simd_maxsz(desc)); \
+}
+
+DO_RSRA(gvec_srsra_b, int8_t)
+DO_RSRA(gvec_srsra_h, int16_t)
+DO_RSRA(gvec_srsra_s, int32_t)
+DO_RSRA(gvec_srsra_d, int64_t)
+
+DO_RSRA(gvec_ursra_b, uint8_t)
+DO_RSRA(gvec_ursra_h, uint16_t)
+DO_RSRA(gvec_ursra_s, uint32_t)
+DO_RSRA(gvec_ursra_d, uint64_t)
+
+#undef DO_RSRA
+
/*
* Convert float16 to float32, raising no exceptions and
* preserving exceptional values, including SNaN.
Create vectorized versions of handle_shri_with_rndacc for shift+round and shift+round+accumulate. Add out-of-line helpers in preparation for longer vector lengths from SVE. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/helper.h | 20 ++ target/arm/translate-a64.h | 9 + target/arm/translate-a64.c | 458 ++++++++++++++++++++++++++++++++++++- target/arm/vec_helper.c | 50 ++++ 4 files changed, 534 insertions(+), 3 deletions(-) -- 2.20.1