target-i386: Compute all flag data inside %cl != 0 test.

The (x << (cl - 1)) quantity is only used if CL != 0.  Move the
computation of that quantity nearer its use.

This avoids the creation of undefined TCG operations when the
constant propagation optimization proves that CL == 0, and thus
CL-1 is outside the range [0-wordsize).

Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: malc <av1474@comtv.ru>
This commit is contained in:
Richard Henderson 2011-09-08 09:23:11 +01:00 committed by malc
parent c82dc29a91
commit 8278604134

View File

@ -1406,70 +1406,84 @@ static void gen_shift_rm_T1(DisasContext *s, int ot, int op1,
{ {
target_ulong mask; target_ulong mask;
int shift_label; int shift_label;
TCGv t0, t1; TCGv t0, t1, t2;
if (ot == OT_QUAD) if (ot == OT_QUAD) {
mask = 0x3f; mask = 0x3f;
else } else {
mask = 0x1f; mask = 0x1f;
}
/* load */ /* load */
if (op1 == OR_TMP0) if (op1 == OR_TMP0) {
gen_op_ld_T0_A0(ot + s->mem_index); gen_op_ld_T0_A0(ot + s->mem_index);
else } else {
gen_op_mov_TN_reg(ot, 0, op1); gen_op_mov_TN_reg(ot, 0, op1);
}
tcg_gen_andi_tl(cpu_T[1], cpu_T[1], mask); t0 = tcg_temp_local_new();
t1 = tcg_temp_local_new();
t2 = tcg_temp_local_new();
tcg_gen_addi_tl(cpu_tmp5, cpu_T[1], -1); tcg_gen_andi_tl(t2, cpu_T[1], mask);
if (is_right) { if (is_right) {
if (is_arith) { if (is_arith) {
gen_exts(ot, cpu_T[0]); gen_exts(ot, cpu_T[0]);
tcg_gen_sar_tl(cpu_T3, cpu_T[0], cpu_tmp5); tcg_gen_mov_tl(t0, cpu_T[0]);
tcg_gen_sar_tl(cpu_T[0], cpu_T[0], cpu_T[1]); tcg_gen_sar_tl(cpu_T[0], cpu_T[0], t2);
} else { } else {
gen_extu(ot, cpu_T[0]); gen_extu(ot, cpu_T[0]);
tcg_gen_shr_tl(cpu_T3, cpu_T[0], cpu_tmp5); tcg_gen_mov_tl(t0, cpu_T[0]);
tcg_gen_shr_tl(cpu_T[0], cpu_T[0], cpu_T[1]); tcg_gen_shr_tl(cpu_T[0], cpu_T[0], t2);
} }
} else { } else {
tcg_gen_shl_tl(cpu_T3, cpu_T[0], cpu_tmp5); tcg_gen_mov_tl(t0, cpu_T[0]);
tcg_gen_shl_tl(cpu_T[0], cpu_T[0], cpu_T[1]); tcg_gen_shl_tl(cpu_T[0], cpu_T[0], t2);
} }
/* store */ /* store */
if (op1 == OR_TMP0) if (op1 == OR_TMP0) {
gen_op_st_T0_A0(ot + s->mem_index); gen_op_st_T0_A0(ot + s->mem_index);
else } else {
gen_op_mov_reg_T0(ot, op1); gen_op_mov_reg_T0(ot, op1);
}
/* update eflags if non zero shift */ /* update eflags if non zero shift */
if (s->cc_op != CC_OP_DYNAMIC) if (s->cc_op != CC_OP_DYNAMIC) {
gen_op_set_cc_op(s->cc_op); gen_op_set_cc_op(s->cc_op);
}
/* XXX: inefficient */ tcg_gen_mov_tl(t1, cpu_T[0]);
t0 = tcg_temp_local_new();
t1 = tcg_temp_local_new();
tcg_gen_mov_tl(t0, cpu_T[0]);
tcg_gen_mov_tl(t1, cpu_T3);
shift_label = gen_new_label(); shift_label = gen_new_label();
tcg_gen_brcondi_tl(TCG_COND_EQ, cpu_T[1], 0, shift_label); tcg_gen_brcondi_tl(TCG_COND_EQ, t2, 0, shift_label);
tcg_gen_mov_tl(cpu_cc_src, t1); tcg_gen_addi_tl(t2, t2, -1);
tcg_gen_mov_tl(cpu_cc_dst, t0); tcg_gen_mov_tl(cpu_cc_dst, t1);
if (is_right)
if (is_right) {
if (is_arith) {
tcg_gen_sar_tl(cpu_cc_src, t0, t2);
} else {
tcg_gen_shr_tl(cpu_cc_src, t0, t2);
}
} else {
tcg_gen_shl_tl(cpu_cc_src, t0, t2);
}
if (is_right) {
tcg_gen_movi_i32(cpu_cc_op, CC_OP_SARB + ot); tcg_gen_movi_i32(cpu_cc_op, CC_OP_SARB + ot);
else } else {
tcg_gen_movi_i32(cpu_cc_op, CC_OP_SHLB + ot); tcg_gen_movi_i32(cpu_cc_op, CC_OP_SHLB + ot);
}
gen_set_label(shift_label); gen_set_label(shift_label);
s->cc_op = CC_OP_DYNAMIC; /* cannot predict flags after */ s->cc_op = CC_OP_DYNAMIC; /* cannot predict flags after */
tcg_temp_free(t0); tcg_temp_free(t0);
tcg_temp_free(t1); tcg_temp_free(t1);
tcg_temp_free(t2);
} }
static void gen_shift_rm_im(DisasContext *s, int ot, int op1, int op2, static void gen_shift_rm_im(DisasContext *s, int ot, int op1, int op2,