On 3/27/23 20:06, Song Gao wrote:
+static void gen_vsat_s(unsigned vece, TCGv_vec t, TCGv_vec a, int64_t imm)
+{
+ TCGv_vec t1;
+ int64_t max = (1l << imm) - 1;
This needed 1ull, but better to just use
max = MAKE_64BIT_MASK(0, imm - 1);
+ int64_t min = ~max;
Extra space.
+ t1 = tcg_temp_new_vec_matching(t);
+ tcg_gen_dupi_vec(vece, t, min);
+ tcg_gen_smax_vec(vece, t, a, t);
Use tcg_constant_vec_matching(t, vece, min) instead of dupi.
Three instances.
+ tcg_gen_dupi_vec(vece, t1, max);
+ tcg_gen_smin_vec(vece, t, t, t1);
+}
+
+static void do_vsat_s(unsigned vece, uint32_t vd_ofs, uint32_t vj_ofs,
+ int64_t imm, uint32_t oprsz, uint32_t maxsz)
+{
+ static const TCGOpcode vecop_list[] = {
+ INDEX_op_smax_vec, INDEX_op_smin_vec, 0
+ };
+ static const GVecGen2i op[4] = {
+ {
+ .fniv = gen_vsat_s,
+ .fnoi = gen_helper_vsat_b,
+ .opt_opc = vecop_list,
+ .vece = MO_8
+ },
+ {
+ .fniv = gen_vsat_s,
+ .fnoi = gen_helper_vsat_h,
+ .opt_opc = vecop_list,
+ .vece = MO_16
+ },
+ {
+ .fniv = gen_vsat_s,
+ .fnoi = gen_helper_vsat_w,
+ .opt_opc = vecop_list,
+ .vece = MO_32
+ },
+ {
+ .fniv = gen_vsat_s,
+ .fnoi = gen_helper_vsat_d,
+ .opt_opc = vecop_list,
+ .vece = MO_64
+ },
+ };
+
+ tcg_gen_gvec_2i(vd_ofs, vj_ofs, oprsz, maxsz, imm, &op[vece]);
Better to expand imm to max here, rather than both inside gen_vsat_s and the runtime
do_vsats_*.
Likewise for the unsigned versions.
r~
