motorhead/qemu
1
0
Fork 0
mirror of https://github.com/Motorhead1991/qemu.git synced 2025-08-06 17:23:56 -06:00
Code Issues Projects Releases Packages Wiki Activity Actions

target/arm: Avoid tcg_const_* in translate-mve.c

Browse source 
All uses are in the context of an accumulator conditionally
having a zero input.  Split the rda variable to rda_{i,o},
and set rda_i to tcg_constant_foo(0) when required.

Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
This commit is contained in:
Richard Henderson 2023-02-25 11:32:36 -10:00
parent d6840b9878
commit 063e6e4527
1 changed files with 29 additions and 25 deletions
Download patch file Download diff file Expand all files Collapse all files

54
target/arm/tcg/translate-mve.c
View file

@ -1150,7 +1150,7 @@ static bool do_long_dual_acc(DisasContext *s, arg_vmlaldav *a,
MVEGenLongDualAccOpFn *fn) MVEGenLongDualAccOpFn *fn)
{ {
TCGv_ptr qn, qm; TCGv_ptr qn, qm;
TCGv_i64 rda; TCGv_i64 rda_i, rda_o;
TCGv_i32 rdalo, rdahi; TCGv_i32 rdalo, rdahi;
if (!dc_isar_feature(aa32_mve, s) || if (!dc_isar_feature(aa32_mve, s) ||
@ -1177,21 +1177,22 @@ static bool do_long_dual_acc(DisasContext *s, arg_vmlaldav *a,
* of an A=0 (no-accumulate) insn which does not execute the first * of an A=0 (no-accumulate) insn which does not execute the first
* beat must start with the current rda value, not 0. * beat must start with the current rda value, not 0.
*/ */
rda_o = tcg_temp_new_i64();
if (a->a || mve_skip_first_beat(s)) { if (a->a || mve_skip_first_beat(s)) {
rda = tcg_temp_new_i64(); rda_i = rda_o;
rdalo = load_reg(s, a->rdalo); rdalo = load_reg(s, a->rdalo);
rdahi = load_reg(s, a->rdahi); rdahi = load_reg(s, a->rdahi);
tcg_gen_concat_i32_i64(rda, rdalo, rdahi); tcg_gen_concat_i32_i64(rda_i, rdalo, rdahi);
} else { } else {
rda = tcg_const_i64(0); rda_i = tcg_constant_i64(0);
} }
fn(rda, cpu_env, qn, qm, rda); fn(rda_o, cpu_env, qn, qm, rda_i);
rdalo = tcg_temp_new_i32(); rdalo = tcg_temp_new_i32();
rdahi = tcg_temp_new_i32(); rdahi = tcg_temp_new_i32();
tcg_gen_extrl_i64_i32(rdalo, rda); tcg_gen_extrl_i64_i32(rdalo, rda_o);
tcg_gen_extrh_i64_i32(rdahi, rda); tcg_gen_extrh_i64_i32(rdahi, rda_o);
store_reg(s, a->rdalo, rdalo); store_reg(s, a->rdalo, rdalo);
store_reg(s, a->rdahi, rdahi); store_reg(s, a->rdahi, rdahi);
mve_update_eci(s); mve_update_eci(s);
@ -1258,7 +1259,7 @@ static bool trans_VRMLSLDAVH(DisasContext *s, arg_vmlaldav *a)
static bool do_dual_acc(DisasContext *s, arg_vmladav *a, MVEGenDualAccOpFn *fn) static bool do_dual_acc(DisasContext *s, arg_vmladav *a, MVEGenDualAccOpFn *fn)
{ {
TCGv_ptr qn, qm; TCGv_ptr qn, qm;
TCGv_i32 rda; TCGv_i32 rda_i, rda_o;
if (!dc_isar_feature(aa32_mve, s) || if (!dc_isar_feature(aa32_mve, s) ||
!mve_check_qreg_bank(s, a->qn) || !mve_check_qreg_bank(s, a->qn) ||
@ -1278,13 +1279,14 @@ static bool do_dual_acc(DisasContext *s, arg_vmladav *a, MVEGenDualAccOpFn *fn)
* beat must start with the current rda value, not 0. * beat must start with the current rda value, not 0.
*/ */
if (a->a || mve_skip_first_beat(s)) { if (a->a || mve_skip_first_beat(s)) {
rda = load_reg(s, a->rda); rda_o = rda_i = load_reg(s, a->rda);
} else { } else {
rda = tcg_const_i32(0); rda_i = tcg_constant_i32(0);
rda_o = tcg_temp_new_i32();
} }
fn(rda, cpu_env, qn, qm, rda); fn(rda_o, cpu_env, qn, qm, rda_i);
store_reg(s, a->rda, rda); store_reg(s, a->rda, rda_o);
mve_update_eci(s); mve_update_eci(s);
return true; return true;
@ -1396,7 +1398,7 @@ static bool trans_VADDV(DisasContext *s, arg_VADDV *a)
{ NULL, NULL } { NULL, NULL }
}; };
TCGv_ptr qm; TCGv_ptr qm;
TCGv_i32 rda; TCGv_i32 rda_i, rda_o;
if (!dc_isar_feature(aa32_mve, s) || if (!dc_isar_feature(aa32_mve, s) ||
a->size == 3) { a->size == 3) {
@ -1413,15 +1415,16 @@ static bool trans_VADDV(DisasContext *s, arg_VADDV *a)
*/ */
if (a->a || mve_skip_first_beat(s)) { if (a->a || mve_skip_first_beat(s)) {
/* Accumulate input from Rda */ /* Accumulate input from Rda */
rda = load_reg(s, a->rda); rda_o = rda_i = load_reg(s, a->rda);
} else { } else {
/* Accumulate starting at zero */ /* Accumulate starting at zero */
rda = tcg_const_i32(0); rda_i = tcg_constant_i32(0);
rda_o = tcg_temp_new_i32();
} }
qm = mve_qreg_ptr(a->qm); qm = mve_qreg_ptr(a->qm);
fns[a->size][a->u](rda, cpu_env, qm, rda); fns[a->size][a->u](rda_o, cpu_env, qm, rda_i);
store_reg(s, a->rda, rda); store_reg(s, a->rda, rda_o);
mve_update_eci(s); mve_update_eci(s);
return true; return true;
@ -1436,7 +1439,7 @@ static bool trans_VADDLV(DisasContext *s, arg_VADDLV *a)
* No need to check Qm's bank: it is only 3 bits in decode. * No need to check Qm's bank: it is only 3 bits in decode.
*/ */
TCGv_ptr qm; TCGv_ptr qm;
TCGv_i64 rda; TCGv_i64 rda_i, rda_o;
TCGv_i32 rdalo, rdahi; TCGv_i32 rdalo, rdahi;
if (!dc_isar_feature(aa32_mve, s)) { if (!dc_isar_feature(aa32_mve, s)) {
@ -1458,28 +1461,29 @@ static bool trans_VADDLV(DisasContext *s, arg_VADDLV *a)
* of an A=0 (no-accumulate) insn which does not execute the first * of an A=0 (no-accumulate) insn which does not execute the first
* beat must start with the current value of RdaHi:RdaLo, not zero. * beat must start with the current value of RdaHi:RdaLo, not zero.
*/ */
rda_o = tcg_temp_new_i64();
if (a->a || mve_skip_first_beat(s)) { if (a->a || mve_skip_first_beat(s)) {
/* Accumulate input from RdaHi:RdaLo */ /* Accumulate input from RdaHi:RdaLo */
rda = tcg_temp_new_i64(); rda_i = rda_o;
rdalo = load_reg(s, a->rdalo); rdalo = load_reg(s, a->rdalo);
rdahi = load_reg(s, a->rdahi); rdahi = load_reg(s, a->rdahi);
tcg_gen_concat_i32_i64(rda, rdalo, rdahi); tcg_gen_concat_i32_i64(rda_i, rdalo, rdahi);
} else { } else {
/* Accumulate starting at zero */ /* Accumulate starting at zero */
rda = tcg_const_i64(0); rda_i = tcg_constant_i64(0);
} }
qm = mve_qreg_ptr(a->qm); qm = mve_qreg_ptr(a->qm);
if (a->u) { if (a->u) {
gen_helper_mve_vaddlv_u(rda, cpu_env, qm, rda); gen_helper_mve_vaddlv_u(rda_o, cpu_env, qm, rda_i);
} else { } else {
gen_helper_mve_vaddlv_s(rda, cpu_env, qm, rda); gen_helper_mve_vaddlv_s(rda_o, cpu_env, qm, rda_i);
} }
rdalo = tcg_temp_new_i32(); rdalo = tcg_temp_new_i32();
rdahi = tcg_temp_new_i32(); rdahi = tcg_temp_new_i32();
tcg_gen_extrl_i64_i32(rdalo, rda); tcg_gen_extrl_i64_i32(rdalo, rda_o);
tcg_gen_extrh_i64_i32(rdahi, rda); tcg_gen_extrh_i64_i32(rdahi, rda_o);
store_reg(s, a->rdalo, rdalo); store_reg(s, a->rdalo, rdalo);
store_reg(s, a->rdahi, rdahi); store_reg(s, a->rdahi, rdahi);
mve_update_eci(s); mve_update_eci(s);