tcg/aarch64: Use tcg_constant_vec with tcg vec expanders

Improve rotrv_vec to reduce "t1 = -v2, t2 = t1 + c" to
"t1 = -v2, t2 = c - v2".  This avoids a serial dependency
between t1 and t2.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
This commit is contained in:
Richard Henderson 2020-09-08 00:47:31 +00:00
parent 44aa59a099
commit 10061ffe56
1 changed files with 5 additions and 5 deletions

View File

@ -2516,7 +2516,7 @@ void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece,
TCGArg a0, ...) TCGArg a0, ...)
{ {
va_list va; va_list va;
TCGv_vec v0, v1, v2, t1, t2; TCGv_vec v0, v1, v2, t1, t2, c1;
TCGArg a2; TCGArg a2;
va_start(va, a0); va_start(va, a0);
@ -2548,8 +2548,8 @@ void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece,
case INDEX_op_rotlv_vec: case INDEX_op_rotlv_vec:
t1 = tcg_temp_new_vec(type); t1 = tcg_temp_new_vec(type);
tcg_gen_dupi_vec(vece, t1, 8 << vece); c1 = tcg_constant_vec(type, vece, 8 << vece);
tcg_gen_sub_vec(vece, t1, v2, t1); tcg_gen_sub_vec(vece, t1, v2, c1);
/* Right shifts are negative left shifts for AArch64. */ /* Right shifts are negative left shifts for AArch64. */
vec_gen_3(INDEX_op_shlv_vec, type, vece, tcgv_vec_arg(t1), vec_gen_3(INDEX_op_shlv_vec, type, vece, tcgv_vec_arg(t1),
tcgv_vec_arg(v1), tcgv_vec_arg(t1)); tcgv_vec_arg(v1), tcgv_vec_arg(t1));
@ -2562,9 +2562,9 @@ void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece,
case INDEX_op_rotrv_vec: case INDEX_op_rotrv_vec:
t1 = tcg_temp_new_vec(type); t1 = tcg_temp_new_vec(type);
t2 = tcg_temp_new_vec(type); t2 = tcg_temp_new_vec(type);
c1 = tcg_constant_vec(type, vece, 8 << vece);
tcg_gen_neg_vec(vece, t1, v2); tcg_gen_neg_vec(vece, t1, v2);
tcg_gen_dupi_vec(vece, t2, 8 << vece); tcg_gen_sub_vec(vece, t2, c1, v2);
tcg_gen_add_vec(vece, t2, t1, t2);
/* Right shifts are negative left shifts for AArch64. */ /* Right shifts are negative left shifts for AArch64. */
vec_gen_3(INDEX_op_shlv_vec, type, vece, tcgv_vec_arg(t1), vec_gen_3(INDEX_op_shlv_vec, type, vece, tcgv_vec_arg(t1),
tcgv_vec_arg(v1), tcgv_vec_arg(t1)); tcgv_vec_arg(v1), tcgv_vec_arg(t1));