Impaczus/rpcs3
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

SPU LLVM: Fix LSA masking for PUTLLC16, disable RTIME checks

Browse Source
This commit is contained in:
Elad Ashkenazi
2024-06-27 10:26:20 +03:00
parent 908082e7c3
commit a9d53e98de
2 changed files with 13 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files
rpcs3/Emu/Cell/SPUCommonRecompiler.cpp
+10 -4
View File
@@ -5689,9 +5689,15 @@ spu_program spu_recompiler_base::analyse(const be_t<u32>* ls, u32 entry_point, s
{ {
case MFC_GETLLAR_CMD: case MFC_GETLLAR_CMD:
{ {
// Get LSA and apply mask for GETLLAR
// TODO: Simplify this to be a value returning function
auto old_lsa = get_reg(s_reg_mfc_lsa);
inherit_const_mask_value(s_reg_mfc_lsa, old_lsa, 0, ~SPU_LS_MASK_128);
// Restore LSA
auto lsa = get_reg(s_reg_mfc_lsa); auto lsa = get_reg(s_reg_mfc_lsa);
inherit_const_mask_value(s_reg_mfc_lsa, lsa, 0, ~SPU_LS_MASK_128); vregs[s_reg_mfc_lsa] = old_lsa;
lsa = get_reg(s_reg_mfc_lsa);
const u32 lsa_pc = atomic16.lsa_last_pc == SPU_LS_SIZE ? bpc : atomic16.lsa_last_pc; const u32 lsa_pc = atomic16.lsa_last_pc == SPU_LS_SIZE ? bpc : atomic16.lsa_last_pc;
if (atomic16.active) if (atomic16.active)
@@ -5743,7 +5749,7 @@ spu_program spu_recompiler_base::analyse(const be_t<u32>* ls, u32 entry_point, s
continue; continue;
} }
if (vregs[s_reg_mfc_lsa].compare_with_mask_indifference(*val, SPU_LS_MASK_1)) if (vregs[s_reg_mfc_lsa].compare_with_mask_indifference(*val, SPU_LS_MASK_16))
{ {
regs[reg_it] = s_reg_mfc_lsa; regs[reg_it] = s_reg_mfc_lsa;
continue; continue;
@@ -5753,7 +5759,7 @@ spu_program spu_recompiler_base::analyse(const be_t<u32>* ls, u32 entry_point, s
{ {
const auto& _reg = vregs[i]; const auto& _reg = vregs[i];
if (_reg == *val) if (_reg.compare_with_mask_indifference(*val, SPU_LS_MASK_16))
{ {
regs[reg_it] = i; regs[reg_it] = i;
break; break;
rpcs3/Emu/Cell/SPULLVMRecompiler.cpp
+3 -3
View File
@@ -1292,7 +1292,7 @@ class spu_llvm_recompiler : public spu_recompiler_base, public cpu_translator
const auto diff = m_ir->CreateZExt(m_ir->CreateSub(dest, _lsa), get_type<u64>()); const auto diff = m_ir->CreateZExt(m_ir->CreateSub(dest, _lsa), get_type<u64>());
const auto _new = m_ir->CreateAlignedLoad(get_type<u128>(), _ptr<u128>(m_lsptr, dest), llvm::MaybeAlign{16}); const auto _new = m_ir->CreateAlignedLoad(get_type<u128>(), _ptr<u128>(m_lsptr, dest), llvm::MaybeAlign{16});
const auto _rdata = m_ir->CreateAlignedLoad(get_type<u128>(), _ptr<u128>(spu_ptr<u8>(&spu_thread::rdata), m_ir->CreateAnd(diff, 0x7f)), llvm::MaybeAlign{16}); const auto _rdata = m_ir->CreateAlignedLoad(get_type<u128>(), _ptr<u128>(spu_ptr<u8>(&spu_thread::rdata), m_ir->CreateAnd(diff, 0x70)), llvm::MaybeAlign{16});
const bool is_accurate_op = !!g_cfg.core.spu_accurate_reservations; const bool is_accurate_op = !!g_cfg.core.spu_accurate_reservations;
@@ -1360,7 +1360,7 @@ class spu_llvm_recompiler : public spu_recompiler_base, public cpu_translator
llvm::Value* old_val{}; llvm::Value* old_val{};
if (is_accurate_op) if (true || is_accurate_op)
{ {
old_val = m_ir->CreateLoad(get_type<u64>(), spu_ptr<u64>(&spu_thread::rtime)); old_val = m_ir->CreateLoad(get_type<u64>(), spu_ptr<u64>(&spu_thread::rtime));
} }
@@ -1373,7 +1373,7 @@ class spu_llvm_recompiler : public spu_recompiler_base, public cpu_translator
const auto cmp_res2 = m_ir->CreateAtomicCmpXchg(rptr2, old_val, m_ir->CreateAdd(old_val, m_ir->getInt64(128)), llvm::MaybeAlign{16}, llvm::AtomicOrdering::SequentiallyConsistent, llvm::AtomicOrdering::SequentiallyConsistent); const auto cmp_res2 = m_ir->CreateAtomicCmpXchg(rptr2, old_val, m_ir->CreateAdd(old_val, m_ir->getInt64(128)), llvm::MaybeAlign{16}, llvm::AtomicOrdering::SequentiallyConsistent, llvm::AtomicOrdering::SequentiallyConsistent);
if (is_accurate_op) if (true || is_accurate_op)
{ {
m_ir->CreateCondBr(m_ir->CreateExtractValue(cmp_res2, 1), _success, _fail); m_ir->CreateCondBr(m_ir->CreateExtractValue(cmp_res2, 1), _success, _fail);
} }