diff options
author | Richard Henderson <richard.henderson@linaro.org> | 2023-02-25 11:32:36 -1000 |
---|---|---|
committer | Richard Henderson <richard.henderson@linaro.org> | 2023-03-13 07:03:39 -0700 |
commit | 063e6e4527fb826e81a1732eed3716a4251e263e (patch) | |
tree | c611f99887bffbae2ab6720a374e9615c0372b29 /target/arm/tcg/translate-mve.c | |
parent | d6840b98787dc2a5ac2dbf687a944da2f832b02d (diff) |
target/arm: Avoid tcg_const_* in translate-mve.c
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>
Diffstat (limited to 'target/arm/tcg/translate-mve.c')
-rw-r--r-- | target/arm/tcg/translate-mve.c | 54 |
1 files changed, 29 insertions, 25 deletions
diff --git a/target/arm/tcg/translate-mve.c b/target/arm/tcg/translate-mve.c index 9744bf3de0..31fb2110f1 100644 --- a/target/arm/tcg/translate-mve.c +++ b/target/arm/tcg/translate-mve.c @@ -1150,7 +1150,7 @@ static bool do_long_dual_acc(DisasContext *s, arg_vmlaldav *a, MVEGenLongDualAccOpFn *fn) { TCGv_ptr qn, qm; - TCGv_i64 rda; + TCGv_i64 rda_i, rda_o; TCGv_i32 rdalo, rdahi; 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 * beat must start with the current rda value, not 0. */ + rda_o = tcg_temp_new_i64(); if (a->a || mve_skip_first_beat(s)) { - rda = tcg_temp_new_i64(); + rda_i = rda_o; rdalo = load_reg(s, a->rdalo); rdahi = load_reg(s, a->rdahi); - tcg_gen_concat_i32_i64(rda, rdalo, rdahi); + tcg_gen_concat_i32_i64(rda_i, rdalo, rdahi); } 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(); rdahi = tcg_temp_new_i32(); - tcg_gen_extrl_i64_i32(rdalo, rda); - tcg_gen_extrh_i64_i32(rdahi, rda); + tcg_gen_extrl_i64_i32(rdalo, rda_o); + tcg_gen_extrh_i64_i32(rdahi, rda_o); store_reg(s, a->rdalo, rdalo); store_reg(s, a->rdahi, rdahi); 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) { TCGv_ptr qn, qm; - TCGv_i32 rda; + TCGv_i32 rda_i, rda_o; if (!dc_isar_feature(aa32_mve, s) || !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. */ if (a->a || mve_skip_first_beat(s)) { - rda = load_reg(s, a->rda); + rda_o = rda_i = load_reg(s, a->rda); } 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); - store_reg(s, a->rda, rda); + fn(rda_o, cpu_env, qn, qm, rda_i); + store_reg(s, a->rda, rda_o); mve_update_eci(s); return true; @@ -1396,7 +1398,7 @@ static bool trans_VADDV(DisasContext *s, arg_VADDV *a) { NULL, NULL } }; TCGv_ptr qm; - TCGv_i32 rda; + TCGv_i32 rda_i, rda_o; if (!dc_isar_feature(aa32_mve, s) || a->size == 3) { @@ -1413,15 +1415,16 @@ static bool trans_VADDV(DisasContext *s, arg_VADDV *a) */ if (a->a || mve_skip_first_beat(s)) { /* Accumulate input from Rda */ - rda = load_reg(s, a->rda); + rda_o = rda_i = load_reg(s, a->rda); } else { /* 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); - fns[a->size][a->u](rda, cpu_env, qm, rda); - store_reg(s, a->rda, rda); + fns[a->size][a->u](rda_o, cpu_env, qm, rda_i); + store_reg(s, a->rda, rda_o); mve_update_eci(s); 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. */ TCGv_ptr qm; - TCGv_i64 rda; + TCGv_i64 rda_i, rda_o; TCGv_i32 rdalo, rdahi; 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 * 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)) { /* Accumulate input from RdaHi:RdaLo */ - rda = tcg_temp_new_i64(); + rda_i = rda_o; rdalo = load_reg(s, a->rdalo); rdahi = load_reg(s, a->rdahi); - tcg_gen_concat_i32_i64(rda, rdalo, rdahi); + tcg_gen_concat_i32_i64(rda_i, rdalo, rdahi); } else { /* Accumulate starting at zero */ - rda = tcg_const_i64(0); + rda_i = tcg_constant_i64(0); } qm = mve_qreg_ptr(a->qm); 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 { - 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(); rdahi = tcg_temp_new_i32(); - tcg_gen_extrl_i64_i32(rdalo, rda); - tcg_gen_extrh_i64_i32(rdahi, rda); + tcg_gen_extrl_i64_i32(rdalo, rda_o); + tcg_gen_extrh_i64_i32(rdahi, rda_o); store_reg(s, a->rdalo, rdalo); store_reg(s, a->rdahi, rdahi); mve_update_eci(s); |