mirror/rpcs3
1
0
Fork 0
mirror of https://github.com/RPCS3/rpcs3.git synced 2025-02-22 03:40:30 +00:00
Code Issues Packages Projects Releases Wiki Activity

Revert FM(a*re_accurate) shortcut to relaxed

Browse Source
This commit is contained in:
RipleyTom 2024-02-13 18:07:26 +01:00 committed by Elad.Ash
parent 50b1402d66
commit af33c0c69f
1 changed files with 27 additions and 22 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

49
rpcs3/Emu/Cell/SPULLVMRecompiler.cpp
View File

@ -5620,32 +5620,37 @@ public:
const auto [a, b] = get_vrs<f32[4]>(op.ra, op.rb); const auto [a, b] = get_vrs<f32[4]>(op.ra, op.rb);
auto full_fm_accurate = [&](const auto& a, const auto& div) // This causes issues in LBP 1(first platform on first temple level doesn't come down when grabbed)
// Presumably 1/x might result in Zero/NaN when a/x doesn't
if (g_cfg.core.spu_xfloat_accuracy == xfloat_accuracy::relaxed)
{ {
const auto div_result = a / div; auto full_fm_accurate = [&](const auto& a, const auto& div)
const auto result_and = bitcast<u32[4]>(div_result) & 0x7fffffffu; {
const auto result_cmp_inf = sext<s32[4]>(result_and == splat<u32[4]>(0x7F800000u)); const auto div_result = a / div;
const auto result_cmp_nan = sext<s32[4]>(result_and <= splat<u32[4]>(0x7F800000u)); const auto result_and = bitcast<u32[4]>(div_result) & 0x7fffffffu;
const auto result_cmp_inf = sext<s32[4]>(result_and == splat<u32[4]>(0x7F800000u));
const auto result_cmp_nan = sext<s32[4]>(result_and <= splat<u32[4]>(0x7F800000u));
const auto and_mask = bitcast<u32[4]>(result_cmp_nan) & splat<u32[4]>(0xFFFFFFFFu); const auto and_mask = bitcast<u32[4]>(result_cmp_nan) & splat<u32[4]>(0xFFFFFFFFu);
const auto or_mask = bitcast<u32[4]>(result_cmp_inf) & splat<u32[4]>(0xFFFFFFFu); const auto or_mask = bitcast<u32[4]>(result_cmp_inf) & splat<u32[4]>(0xFFFFFFFu);
set_vr(op.rt, bitcast<f32[4]>((bitcast<u32[4]>(div_result) & and_mask) | or_mask)); set_vr(op.rt, bitcast<f32[4]>((bitcast<u32[4]>(div_result) & and_mask) | or_mask));
}; };
// FM(a, re_accurate(div)) // FM(a, re_accurate(div))
if (const auto [ok_re_acc, div, one] = match_expr(b, re_accurate(match<f32[4]>(), match<f32[4]>())); ok_re_acc) if (const auto [ok_re_acc, div, one] = match_expr(b, re_accurate(match<f32[4]>(), match<f32[4]>())); ok_re_acc)
{ {
full_fm_accurate(a, div); full_fm_accurate(a, div);
erase_stores(one, b); erase_stores(one, b);
return; return;
} }
// FM(re_accurate(div), b) // FM(re_accurate(div), b)
if (const auto [ok_re_acc, div, one] = match_expr(a, re_accurate(match<f32[4]>(), match<f32[4]>())); ok_re_acc) if (const auto [ok_re_acc, div, one] = match_expr(a, re_accurate(match<f32[4]>(), match<f32[4]>())); ok_re_acc)
{ {
full_fm_accurate(b, div); full_fm_accurate(b, div);
erase_stores(one, a); erase_stores(one, a);
return; return;
}
} }
set_vr(op.rt, fm(a, b)); set_vr(op.rt, fm(a, b));