yuzu/unicorn
1
0
Fork 0
mirror of https://github.com/yuzu-emu/unicorn.git synced 2024-12-23 09:55:33 +00:00
Code Issues Packages Projects Releases Wiki Activity

target/arm: Simplify do_long_3d and do_2scalar_long

Browse source 
In both cases, we can sink the write-back and perform
the accumulate into the normal destination temps

Backports 9f1a5f93c2dd345dc6c8fe86ed14bf1485056f6e
This commit is contained in:
Richard Henderson 2021-03-02 12:46:51 -05:00 committed by Lioncash
parent 07c2b70234
commit 9263117d47
1 changed files with 9 additions and 14 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

23
qemu/target/arm/translate-neon.inc.c
View file

@ -2069,17 +2069,14 @@ static bool do_long_3d(DisasContext *s, arg_3diff *a,
if (accfn) { if (accfn) {
tmp = tcg_temp_new_i64(tcg_ctx); tmp = tcg_temp_new_i64(tcg_ctx);
read_neon_element64(s, tmp, a->vd, 0, MO_64); read_neon_element64(s, tmp, a->vd, 0, MO_64);
accfn(tcg_ctx, tmp, tmp, rd0); accfn(tcg_ctx, rd0, tmp, rd0);
write_neon_element64(s, tmp, a->vd, 0, MO_64);
read_neon_element64(s, tmp, a->vd, 1, MO_64); read_neon_element64(s, tmp, a->vd, 1, MO_64);
accfn(tcg_ctx, tmp, tmp, rd1); accfn(tcg_ctx, rd1, tmp, rd1);
write_neon_element64(s, tmp, a->vd, 1, MO_64);
tcg_temp_free_i64(tcg_ctx, tmp); tcg_temp_free_i64(tcg_ctx, tmp);
} else {
write_neon_element64(s, rd0, a->vd, 0, MO_64);
write_neon_element64(s, rd1, a->vd, 1, MO_64);
} }
write_neon_element64(s, rd0, a->vd, 0, MO_64);
write_neon_element64(s, rd1, a->vd, 1, MO_64);
tcg_temp_free_i64(tcg_ctx, rd0); tcg_temp_free_i64(tcg_ctx, rd0);
tcg_temp_free_i64(tcg_ctx, rd1); tcg_temp_free_i64(tcg_ctx, rd1);
@ -2709,16 +2706,14 @@ static bool do_2scalar_long(DisasContext *s, arg_2scalar *a,
if (accfn) { if (accfn) {
TCGv_i64 t64 = tcg_temp_new_i64(tcg_ctx); TCGv_i64 t64 = tcg_temp_new_i64(tcg_ctx);
read_neon_element64(s, t64, a->vd, 0, MO_64); read_neon_element64(s, t64, a->vd, 0, MO_64);
accfn(tcg_ctx, t64, t64, rn0_64); accfn(tcg_ctx, rn0_64, t64, rn0_64);
write_neon_element64(s, t64, a->vd, 0, MO_64);
read_neon_element64(s, t64, a->vd, 1, MO_64); read_neon_element64(s, t64, a->vd, 1, MO_64);
accfn(tcg_ctx, t64, t64, rn1_64); accfn(tcg_ctx, rn1_64, t64, rn1_64);
write_neon_element64(s, t64, a->vd, 1, MO_64);
tcg_temp_free_i64(tcg_ctx, t64); tcg_temp_free_i64(tcg_ctx, t64);
} else {
write_neon_element64(s, rn0_64, a->vd, 0, MO_64);
write_neon_element64(s, rn1_64, a->vd, 1, MO_64);
} }
write_neon_element64(s, rn0_64, a->vd, 0, MO_64);
write_neon_element64(s, rn1_64, a->vd, 1, MO_64);
tcg_temp_free_i64(tcg_ctx, rn0_64); tcg_temp_free_i64(tcg_ctx, rn0_64);
tcg_temp_free_i64(tcg_ctx, rn1_64); tcg_temp_free_i64(tcg_ctx, rn1_64);
return true; return true;