ShuriZma
/
rpcs3
mirror of https://github.com/RPCS3/rpcs3.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #1015 from gopalsr83/master 

Miscellaneous fixes and improvements to the PPU LLVM JIT
This commit is contained in:
B1ackDaemon 2015-02-18 21:55:25 +02:00
parent cae85c6247 4272ec7a82
commit 1003c42310
5 changed files with 681 additions and 762 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

338
rpcs3/Emu/Cell/PPULLVMRecompiler.cpp
View File

@ -26,8 +26,10 @@ using namespace ppu_recompiler_llvm;
u64 Compiler::s_rotate_mask[64][64];
bool Compiler::s_rotate_mask_inited = false;
Compiler::Compiler(RecompilationEngine & recompilation_engine, const Executable execute_unknown_function, const Executable execute_unknown_block)
: m_recompilation_engine(recompilation_engine) {
Compiler::Compiler(RecompilationEngine & recompilation_engine, const Executable execute_unknown_function,
const Executable execute_unknown_block, bool (*poll_status_function)(PPUThread * ppu_state))
: m_recompilation_engine(recompilation_engine)
, m_poll_status_function(poll_status_function) {
InitializeNativeTarget();
InitializeNativeTargetAsmPrinter();
InitializeNativeTargetDisassembler();
@ -157,20 +159,21 @@ Executable Compiler::Compile(const std::string & name, const ControlFlowGraph &
}
// Found an empty block
m_state.current_instruction_address = GetAddressFromBasicBlockName(block_i->getName());
m_ir_builder->SetInsertPoint(block_i);
auto exit_instr_i32 = m_ir_builder->CreatePHI(m_ir_builder->getInt32Ty(), 0);
exit_instr_list.push_back(exit_instr_i32);
auto instr_address = GetAddressFromBasicBlockName(block_i->getName());
SetPc(m_ir_builder->getInt32(instr_address));
SetPc(m_ir_builder->getInt32(m_state.current_instruction_address));
if (generate_linkable_exits) {
auto context_i64 = m_ir_builder->CreateZExt(exit_instr_i32, m_ir_builder->getInt64Ty());
context_i64 = m_ir_builder->CreateOr(context_i64, (u64)cfg.function_address << 32);
auto ret_i32 = IndirectCall(instr_address, context_i64, false);
auto ret_i32 = IndirectCall(m_state.current_instruction_address, context_i64, false);
auto cmp_i1 = m_ir_builder->CreateICmpNE(ret_i32, m_ir_builder->getInt32(0));
auto then_bb = GetBasicBlockFromAddress(instr_address, "then");
auto merge_bb = GetBasicBlockFromAddress(instr_address, "merge");
auto then_bb = GetBasicBlockFromAddress(m_state.current_instruction_address, "then_0");
auto merge_bb = GetBasicBlockFromAddress(m_state.current_instruction_address, "merge_0");
m_ir_builder->CreateCondBr(cmp_i1, then_bb, merge_bb);
m_ir_builder->SetInsertPoint(then_bb);
@ -195,8 +198,8 @@ Executable Compiler::Compile(const std::string & name, const ControlFlowGraph &
if (generate_linkable_exits) {
auto cmp_i1 = m_ir_builder->CreateICmpNE(exit_instr_i32, m_ir_builder->getInt32(0));
auto then_bb = GetBasicBlockFromAddress(0xFFFFFFFF, "then");
auto merge_bb = GetBasicBlockFromAddress(0xFFFFFFFF, "merge");
auto then_bb = GetBasicBlockFromAddress(0xFFFFFFFF, "then_0");
auto merge_bb = GetBasicBlockFromAddress(0xFFFFFFFF, "merge_0");
m_ir_builder->CreateCondBr(cmp_i1, then_bb, merge_bb);
m_ir_builder->SetInsertPoint(then_bb);
@ -1996,7 +1999,7 @@ void Compiler::BC(u32 bo, u32 bi, s32 bd, u32 aa, u32 lk) {
}
void Compiler::HACK(u32 index) {
Call<void>("execute_ps3_func_by_index", &execute_ps3_func_by_index, m_state.args[CompileTaskState::Args::State], m_ir_builder->getInt32(index));
Call<void>("execute_ps3_func_by_index", &execute_ps3_func_by_index, m_state.args[CompileTaskState::Args::State], m_ir_builder->getInt32(index));
}
void Compiler::SC(u32 lev) {
@ -2015,6 +2018,15 @@ void Compiler::SC(u32 lev) {
CompilationError(fmt::Format("SC %u", lev));
break;
}
auto ret_i1 = Call<bool>("PollStatus", m_poll_status_function, m_state.args[CompileTaskState::Args::State]);
auto cmp_i1 = m_ir_builder->CreateICmpEQ(ret_i1, m_ir_builder->getInt1(true));
auto then_bb = GetBasicBlockFromAddress(m_state.current_instruction_address, "then_true");
auto merge_bb = GetBasicBlockFromAddress(m_state.current_instruction_address, "merge_true");
m_ir_builder->CreateCondBr(cmp_i1, then_bb, merge_bb);
m_ir_builder->SetInsertPoint(then_bb);
m_ir_builder->CreateRet(m_ir_builder->getInt32(0xFFFFFFFF));
m_ir_builder->SetInsertPoint(merge_bb);
}
void Compiler::B(s32 ll, u32 aa, u32 lk) {
@ -2467,27 +2479,20 @@ void Compiler::MFOCRF(u32 a, u32 rd, u32 crm) {
}
void Compiler::LWARX(u32 rd, u32 ra, u32 rb) {
throw __FUNCTION__;
//auto addr_i64 = GetGpr(rb);
//if (ra) {
// auto ra_i64 = GetGpr(ra);
// addr_i64 = m_ir_builder->CreateAdd(ra_i64, addr_i64);
//}
auto addr_i64 = GetGpr(rb);
if (ra) {
auto ra_i64 = GetGpr(ra);
addr_i64 = m_ir_builder->CreateAdd(ra_i64, addr_i64);
}
//auto resv_addr_i8_ptr = m_ir_builder->CreateConstGEP1_32(m_state.args[CompileTaskState::Args::State], (unsigned int)offsetof(PPUThread, R_ADDR));
//auto resv_addr_i64_ptr = m_ir_builder->CreateBitCast(resv_addr_i8_ptr, m_ir_builder->getInt64Ty()->getPointerTo());
//m_ir_builder->CreateAlignedStore(addr_i64, resv_addr_i64_ptr, 8);
//auto resv_val_i32 = ReadMemory(addr_i64, 32, 4, false, false);
//auto resv_val_i64 = m_ir_builder->CreateZExt(resv_val_i32, m_ir_builder->getInt64Ty());
//auto resv_val_i8_ptr = m_ir_builder->CreateConstGEP1_32(m_state.args[CompileTaskState::Args::State], (unsigned int)offsetof(PPUThread, R_VALUE));
//auto resv_val_i64_ptr = m_ir_builder->CreateBitCast(resv_val_i8_ptr, m_ir_builder->getInt64Ty()->getPointerTo());
//m_ir_builder->CreateAlignedStore(resv_val_i64, resv_val_i64_ptr, 8);
//resv_val_i32 = m_ir_builder->CreateCall(Intrinsic::getDeclaration(m_module, Intrinsic::bswap, m_ir_builder->getInt32Ty()), resv_val_i32);
//resv_val_i64 = m_ir_builder->CreateZExt(resv_val_i32, m_ir_builder->getInt64Ty());
//SetGpr(rd, resv_val_i64);
auto addr_i32 = m_ir_builder->CreateTrunc(addr_i64, m_ir_builder->getInt32Ty());
auto val_i32_ptr = m_ir_builder->CreateAlloca(m_ir_builder->getInt32Ty());
val_i32_ptr->setAlignment(4);
Call<bool>("vm.reservation_acquire_no_cb", vm::reservation_acquire_no_cb, m_ir_builder->CreateBitCast(val_i32_ptr, m_ir_builder->getInt8PtrTy()), addr_i32, m_ir_builder->getInt32(4));
auto val_i32 = (Value *)m_ir_builder->CreateLoad(val_i32_ptr);
val_i32 = m_ir_builder->CreateCall(Intrinsic::getDeclaration(m_module, Intrinsic::bswap, m_ir_builder->getInt32Ty()), val_i32);
auto val_i64 = m_ir_builder->CreateZExt(val_i32, m_ir_builder->getInt64Ty());
SetGpr(rd, val_i64);
}
void Compiler::LDX(u32 rd, u32 ra, u32 rb) {
@ -2745,25 +2750,19 @@ void Compiler::MULHW(u32 rd, u32 ra, u32 rb, bool rc) {
}
void Compiler::LDARX(u32 rd, u32 ra, u32 rb) {
throw __FUNCTION__;
auto addr_i64 = GetGpr(rb);
if (ra) {
auto ra_i64 = GetGpr(ra);
addr_i64 = m_ir_builder->CreateAdd(ra_i64, addr_i64);
}
//auto addr_i64 = GetGpr(rb);
//if (ra) {
// auto ra_i64 = GetGpr(ra);
// addr_i64 = m_ir_builder->CreateAdd(ra_i64, addr_i64);
//}
//auto resv_addr_i8_ptr = m_ir_builder->CreateConstGEP1_32(m_state.args[CompileTaskState::Args::State], (unsigned int)offsetof(PPUThread, R_ADDR));
//auto resv_addr_i64_ptr = m_ir_builder->CreateBitCast(resv_addr_i8_ptr, m_ir_builder->getInt64Ty()->getPointerTo());
//m_ir_builder->CreateAlignedStore(addr_i64, resv_addr_i64_ptr, 8);
//auto resv_val_i64 = ReadMemory(addr_i64, 64, 8, false);
//auto resv_val_i8_ptr = m_ir_builder->CreateConstGEP1_32(m_state.args[CompileTaskState::Args::State], (unsigned int)offsetof(PPUThread, R_VALUE));
//auto resv_val_i64_ptr = m_ir_builder->CreateBitCast(resv_val_i8_ptr, m_ir_builder->getInt64Ty()->getPointerTo());
//m_ir_builder->CreateAlignedStore(resv_val_i64, resv_val_i64_ptr, 8);
//resv_val_i64 = m_ir_builder->CreateCall(Intrinsic::getDeclaration(m_module, Intrinsic::bswap, m_ir_builder->getInt64Ty()), resv_val_i64);
//SetGpr(rd, resv_val_i64);
auto addr_i32 = m_ir_builder->CreateTrunc(addr_i64, m_ir_builder->getInt32Ty());
auto val_i64_ptr = m_ir_builder->CreateAlloca(m_ir_builder->getInt64Ty());
val_i64_ptr->setAlignment(8);
Call<bool>("vm.reservation_acquire_no_cb", vm::reservation_acquire_no_cb, m_ir_builder->CreateBitCast(val_i64_ptr, m_ir_builder->getInt8PtrTy()), addr_i32, m_ir_builder->getInt32(8));
auto val_i64 = (Value *)m_ir_builder->CreateLoad(val_i64_ptr);
val_i64 = m_ir_builder->CreateCall(Intrinsic::getDeclaration(m_module, Intrinsic::bswap, m_ir_builder->getInt64Ty()), val_i64);
SetGpr(rd, val_i64);
}
void Compiler::DCBF(u32 ra, u32 rb) {
@ -2927,47 +2926,23 @@ void Compiler::STDX(u32 rs, u32 ra, u32 rb) {
}
void Compiler::STWCX_(u32 rs, u32 ra, u32 rb) {
throw __FUNCTION__;
auto addr_i64 = GetGpr(rb);
if (ra) {
auto ra_i64 = GetGpr(ra);
addr_i64 = m_ir_builder->CreateAdd(ra_i64, addr_i64);
}
//auto addr_i64 = GetGpr(rb);
//if (ra) {
// auto ra_i64 = GetGpr(ra);
// addr_i64 = m_ir_builder->CreateAdd(ra_i64, addr_i64);
//}
auto addr_i32 = m_ir_builder->CreateTrunc(addr_i64, m_ir_builder->getInt32Ty());
auto rs_i32 = GetGpr(rs, 32);
rs_i32 = m_ir_builder->CreateCall(Intrinsic::getDeclaration(m_module, Intrinsic::bswap, m_ir_builder->getInt32Ty()), rs_i32);
auto rs_i32_ptr = m_ir_builder->CreateAlloca(m_ir_builder->getInt32Ty());
rs_i32_ptr->setAlignment(4);
m_ir_builder->CreateStore(rs_i32, rs_i32_ptr);
auto success_i1 = Call<bool>("vm.reservation_update", vm::reservation_update, addr_i32, m_ir_builder->CreateBitCast(rs_i32_ptr, m_ir_builder->getInt8PtrTy()), m_ir_builder->getInt32(4));
//auto resv_addr_i8_ptr = m_ir_builder->CreateConstGEP1_32(m_state.args[CompileTaskState::Args::State], (unsigned int)offsetof(PPUThread, R_ADDR));
//auto resv_addr_i64_ptr = m_ir_builder->CreateBitCast(resv_addr_i8_ptr, m_ir_builder->getInt64Ty()->getPointerTo());
//auto resv_addr_i64 = (Value *)m_ir_builder->CreateAlignedLoad(resv_addr_i64_ptr, 8);
//auto cmp_i1 = m_ir_builder->CreateICmpEQ(addr_i64, resv_addr_i64);
//auto then_bb = GetBasicBlockFromAddress(m_state.current_instruction_address, "then");
//auto else_bb = GetBasicBlockFromAddress(m_state.current_instruction_address, "else");
//auto merge_bb = GetBasicBlockFromAddress(m_state.current_instruction_address, "merge");
//m_ir_builder->CreateCondBr(cmp_i1, then_bb, else_bb);
//m_ir_builder->SetInsertPoint(then_bb);
//auto rs_i32 = GetGpr(rs, 32);
//rs_i32 = m_ir_builder->CreateCall(Intrinsic::getDeclaration(m_module, Intrinsic::bswap, rs_i32->getType()), rs_i32);
//resv_addr_i64 = m_ir_builder->CreateAdd(resv_addr_i64, m_ir_builder->getInt64((u64)vm::get_ptr<u8>(0)));
//auto resv_addr_val_i32_ptr = m_ir_builder->CreateIntToPtr(resv_addr_i64, m_ir_builder->getInt32Ty()->getPointerTo());
//auto resv_val_i8_ptr = m_ir_builder->CreateConstGEP1_32(m_state.args[CompileTaskState::Args::State], (unsigned int)offsetof(PPUThread, R_VALUE));
//auto resv_val_i32_ptr = m_ir_builder->CreateBitCast(resv_val_i8_ptr, m_ir_builder->getInt32Ty()->getPointerTo());
//auto resv_val_i32 = m_ir_builder->CreateAlignedLoad(resv_val_i32_ptr, 8);
//auto res_s = m_ir_builder->CreateAtomicCmpXchg(resv_addr_val_i32_ptr, resv_val_i32, rs_i32, AtomicOrdering::AcquireRelease, AtomicOrdering::Monotonic);
//auto success_i1 = m_ir_builder->CreateExtractValue(res_s, {1});
//auto cr_i32 = GetCr();
//cr_i32 = SetBit(cr_i32, 2, success_i1);
//SetCr(cr_i32);
//m_ir_builder->CreateAlignedStore(m_ir_builder->getInt64(0), resv_addr_i64_ptr, 8);
//m_ir_builder->CreateBr(merge_bb);
//m_ir_builder->SetInsertPoint(else_bb);
//cr_i32 = GetCr();
//cr_i32 = ClrBit(cr_i32, 2);
//SetCr(cr_i32);
//m_ir_builder->CreateBr(merge_bb);
//m_ir_builder->SetInsertPoint(merge_bb);
auto cr_i32 = GetCr();
cr_i32 = SetBit(cr_i32, 2, success_i1);
SetCr(cr_i32);
}
void Compiler::STWX(u32 rs, u32 ra, u32 rb) {
@ -3070,47 +3045,23 @@ void Compiler::SUBFZE(u32 rd, u32 ra, u32 oe, bool rc) {
}
void Compiler::STDCX_(u32 rs, u32 ra, u32 rb) {
throw __FUNCTION__;
auto addr_i64 = GetGpr(rb);
if (ra) {
auto ra_i64 = GetGpr(ra);
addr_i64 = m_ir_builder->CreateAdd(ra_i64, addr_i64);
}
//auto addr_i64 = GetGpr(rb);
//if (ra) {
// auto ra_i64 = GetGpr(ra);
// addr_i64 = m_ir_builder->CreateAdd(ra_i64, addr_i64);
//}
auto addr_i32 = m_ir_builder->CreateTrunc(addr_i64, m_ir_builder->getInt32Ty());
auto rs_i64 = GetGpr(rs);
rs_i64 = m_ir_builder->CreateCall(Intrinsic::getDeclaration(m_module, Intrinsic::bswap, m_ir_builder->getInt64Ty()), rs_i64);
auto rs_i64_ptr = m_ir_builder->CreateAlloca(m_ir_builder->getInt64Ty());
rs_i64_ptr->setAlignment(8);
m_ir_builder->CreateStore(rs_i64, rs_i64_ptr);
auto success_i1 = Call<bool>("vm.reservation_update", vm::reservation_update, addr_i32, m_ir_builder->CreateBitCast(rs_i64_ptr, m_ir_builder->getInt8PtrTy()), m_ir_builder->getInt32(8));
//auto resv_addr_i8_ptr = m_ir_builder->CreateConstGEP1_32(m_state.args[CompileTaskState::Args::State], (unsigned int)offsetof(PPUThread, R_ADDR));
//auto resv_addr_i64_ptr = m_ir_builder->CreateBitCast(resv_addr_i8_ptr, m_ir_builder->getInt64Ty()->getPointerTo());
//auto resv_addr_i64 = (Value *)m_ir_builder->CreateAlignedLoad(resv_addr_i64_ptr, 8);
//auto cmp_i1 = m_ir_builder->CreateICmpEQ(addr_i64, resv_addr_i64);
//auto then_bb = GetBasicBlockFromAddress(m_state.current_instruction_address, "then");
//auto else_bb = GetBasicBlockFromAddress(m_state.current_instruction_address, "else");
//auto merge_bb = GetBasicBlockFromAddress(m_state.current_instruction_address, "merge");
//m_ir_builder->CreateCondBr(cmp_i1, then_bb, else_bb);
//m_ir_builder->SetInsertPoint(then_bb);
//auto rs_i64 = GetGpr(rs, 64);
//rs_i64 = m_ir_builder->CreateCall(Intrinsic::getDeclaration(m_module, Intrinsic::bswap, rs_i64->getType()), rs_i64);
//resv_addr_i64 = m_ir_builder->CreateAdd(resv_addr_i64, m_ir_builder->getInt64((u64)vm::get_ptr<u8>(0)));
//auto resv_addr_val_i64_ptr = m_ir_builder->CreateIntToPtr(resv_addr_i64, m_ir_builder->getInt64Ty()->getPointerTo());
//auto resv_val_i8_ptr = m_ir_builder->CreateConstGEP1_32(m_state.args[CompileTaskState::Args::State], (unsigned int)offsetof(PPUThread, R_VALUE));
//auto resv_val_i64_ptr = m_ir_builder->CreateBitCast(resv_val_i8_ptr, m_ir_builder->getInt64Ty()->getPointerTo());
//auto resv_val_i64 = m_ir_builder->CreateAlignedLoad(resv_val_i64_ptr, 8);
//auto res_s = m_ir_builder->CreateAtomicCmpXchg(resv_addr_val_i64_ptr, resv_val_i64, rs_i64, AtomicOrdering::AcquireRelease, AtomicOrdering::Monotonic);
//auto success_i1 = m_ir_builder->CreateExtractValue(res_s, {1});
//auto cr_i32 = GetCr();
//cr_i32 = SetBit(cr_i32, 2, success_i1);
//SetCr(cr_i32);
//m_ir_builder->CreateAlignedStore(m_ir_builder->getInt64(0), resv_addr_i64_ptr, 8);
//m_ir_builder->CreateBr(merge_bb);
//m_ir_builder->SetInsertPoint(else_bb);
//cr_i32 = GetCr();
//cr_i32 = ClrBit(cr_i32, 2);
//SetCr(cr_i32);
//m_ir_builder->CreateBr(merge_bb);
//m_ir_builder->SetInsertPoint(merge_bb);
auto cr_i32 = GetCr();
cr_i32 = SetBit(cr_i32, 2, success_i1);
SetCr(cr_i32);
}
void Compiler::STBX(u32 rs, u32 ra, u32 rb) {
@ -3275,7 +3226,7 @@ void Compiler::EQV(u32 ra, u32 rs, u32 rb, bool rc) {
}
void Compiler::ECIWX(u32 rd, u32 ra, u32 rb) {
throw __FUNCTION__;
CompilationError("ECIWX");
//auto addr_i64 = GetGpr(rb);
//if (ra) {
// auto ra_i64 = GetGpr(ra);
@ -3435,7 +3386,7 @@ void Compiler::ORC(u32 ra, u32 rs, u32 rb, bool rc) {
}
void Compiler::ECOWX(u32 rs, u32 ra, u32 rb) {
throw __FUNCTION__;
CompilationError("ECOWX");
//auto addr_i64 = GetGpr(rb);
//if (ra) {
// auto ra_i64 = GetGpr(ra);
@ -5530,92 +5481,26 @@ void Compiler::CreateBranch(llvm::Value * cmp_i1, llvm::Value * target_i32, bool
}
Value * Compiler::ReadMemory(Value * addr_i64, u32 bits, u32 alignment, bool bswap, bool could_be_mmio) {
if (bits != 32 || could_be_mmio == false) {
auto eaddr_i64 = m_ir_builder->CreateAdd(addr_i64, m_ir_builder->getInt64((u64)vm::get_ptr<u8>(0)));
auto eaddr_ix_ptr = m_ir_builder->CreateIntToPtr(eaddr_i64, m_ir_builder->getIntNTy(bits)->getPointerTo());
auto val_ix = (Value *)m_ir_builder->CreateLoad(eaddr_ix_ptr, alignment);
if (bits > 8 && bswap) {
val_ix = m_ir_builder->CreateCall(Intrinsic::getDeclaration(m_module, Intrinsic::bswap, m_ir_builder->getIntNTy(bits)), val_ix);
}
return val_ix;
} else {
static u32 next_basic_block_id = 0;
next_basic_block_id++;
auto cmp_i1 = m_ir_builder->CreateICmpULT(addr_i64, m_ir_builder->getInt64(RAW_SPU_BASE_ADDR));
auto then_bb = GetBasicBlockFromAddress(m_state.current_instruction_address, fmt::Format("then_%u", next_basic_block_id));
auto else_bb = GetBasicBlockFromAddress(m_state.current_instruction_address, fmt::Format("else_%u", next_basic_block_id));
auto merge_bb = GetBasicBlockFromAddress(m_state.current_instruction_address, fmt::Format("merge_%u", next_basic_block_id));
m_ir_builder->CreateCondBr(cmp_i1, then_bb, else_bb);
m_ir_builder->SetInsertPoint(then_bb);
auto eaddr_i64 = m_ir_builder->CreateAdd(addr_i64, m_ir_builder->getInt64((u64)vm::get_ptr<u8>(0)));
auto eaddr_i32_ptr = m_ir_builder->CreateIntToPtr(eaddr_i64, m_ir_builder->getInt32Ty()->getPointerTo());
auto val_then_i32 = (Value *)m_ir_builder->CreateAlignedLoad(eaddr_i32_ptr, alignment);
if (bswap) {
val_then_i32 = m_ir_builder->CreateCall(Intrinsic::getDeclaration(m_module, Intrinsic::bswap, m_ir_builder->getInt32Ty()), val_then_i32);
}
m_ir_builder->CreateBr(merge_bb);
m_ir_builder->SetInsertPoint(else_bb);
auto val_else_i32 = Call<u32>("vm.read32", (u32(*)(u64))vm::read32, addr_i64);
if (!bswap) {
val_else_i32 = m_ir_builder->CreateCall(Intrinsic::getDeclaration(m_module, Intrinsic::bswap, m_ir_builder->getInt32Ty()), val_else_i32);
}
m_ir_builder->CreateBr(merge_bb);
m_ir_builder->SetInsertPoint(merge_bb);
auto phi = m_ir_builder->CreatePHI(m_ir_builder->getInt32Ty(), 2);
phi->addIncoming(val_then_i32, then_bb);
phi->addIncoming(val_else_i32, else_bb);
return phi;
addr_i64 = m_ir_builder->CreateAnd(addr_i64, 0xFFFFFFFF);
auto eaddr_i64 = m_ir_builder->CreateAdd(addr_i64, m_ir_builder->getInt64((u64)vm::get_ptr<u8>(0)));
auto eaddr_ix_ptr = m_ir_builder->CreateIntToPtr(eaddr_i64, m_ir_builder->getIntNTy(bits)->getPointerTo());
auto val_ix = (Value *)m_ir_builder->CreateLoad(eaddr_ix_ptr, alignment);
if (bits > 8 && bswap) {
val_ix = m_ir_builder->CreateCall(Intrinsic::getDeclaration(m_module, Intrinsic::bswap, m_ir_builder->getIntNTy(bits)), val_ix);
}
return val_ix;
}
void Compiler::WriteMemory(Value * addr_i64, Value * val_ix, u32 alignment, bool bswap, bool could_be_mmio) {
addr_i64 = m_ir_builder->CreateAnd(addr_i64, 0xFFFFFFFF);
if (val_ix->getType()->getIntegerBitWidth() != 32 || could_be_mmio == false) {
if (val_ix->getType()->getIntegerBitWidth() > 8 && bswap) {
val_ix = m_ir_builder->CreateCall(Intrinsic::getDeclaration(m_module, Intrinsic::bswap, val_ix->getType()), val_ix);
}
auto eaddr_i64 = m_ir_builder->CreateAdd(addr_i64, m_ir_builder->getInt64((u64)vm::get_ptr<u8>(0)));
auto eaddr_ix_ptr = m_ir_builder->CreateIntToPtr(eaddr_i64, val_ix->getType()->getPointerTo());
m_ir_builder->CreateAlignedStore(val_ix, eaddr_ix_ptr, alignment);
} else {
static u32 next_basic_block_id;
next_basic_block_id++;
auto cmp_i1 = m_ir_builder->CreateICmpULT(addr_i64, m_ir_builder->getInt64(RAW_SPU_BASE_ADDR));
auto then_bb = GetBasicBlockFromAddress(m_state.current_instruction_address, fmt::Format("then_%u", next_basic_block_id));
auto else_bb = GetBasicBlockFromAddress(m_state.current_instruction_address, fmt::Format("else_%u", next_basic_block_id));
auto merge_bb = GetBasicBlockFromAddress(m_state.current_instruction_address, fmt::Format("merge_%u", next_basic_block_id));
m_ir_builder->CreateCondBr(cmp_i1, then_bb, else_bb);
m_ir_builder->SetInsertPoint(then_bb);
Value * val_then_i32 = val_ix;
if (bswap) {
val_then_i32 = m_ir_builder->CreateCall(Intrinsic::getDeclaration(m_module, Intrinsic::bswap, m_ir_builder->getInt32Ty()), val_then_i32);
}
auto eaddr_i64 = m_ir_builder->CreateAdd(addr_i64, m_ir_builder->getInt64((u64)vm::get_ptr<u8>(0)));
auto eaddr_i32_ptr = m_ir_builder->CreateIntToPtr(eaddr_i64, m_ir_builder->getInt32Ty()->getPointerTo());
m_ir_builder->CreateAlignedStore(val_then_i32, eaddr_i32_ptr, alignment);
m_ir_builder->CreateBr(merge_bb);
m_ir_builder->SetInsertPoint(else_bb);
Value * val_else_i32 = val_ix;
if (!bswap) {
val_else_i32 = m_ir_builder->CreateCall(Intrinsic::getDeclaration(m_module, Intrinsic::bswap, m_ir_builder->getInt32Ty()), val_else_i32);
}
Call<void>("vm.write32", (void(*)(u32, u32))vm::write32, addr_i64, val_else_i32);
m_ir_builder->CreateBr(merge_bb);
m_ir_builder->SetInsertPoint(merge_bb);
if (val_ix->getType()->getIntegerBitWidth() > 8 && bswap) {
val_ix = m_ir_builder->CreateCall(Intrinsic::getDeclaration(m_module, Intrinsic::bswap, val_ix->getType()), val_ix);
}
addr_i64 = m_ir_builder->CreateAnd(addr_i64, 0xFFFFFFFF);
auto eaddr_i64 = m_ir_builder->CreateAdd(addr_i64, m_ir_builder->getInt64((u64)vm::get_ptr<u8>(0)));
auto eaddr_ix_ptr = m_ir_builder->CreateIntToPtr(eaddr_i64, val_ix->getType()->getPointerTo());
m_ir_builder->CreateAlignedStore(val_ix, eaddr_ix_ptr, alignment);
}
template<class T>
@ -5634,6 +5519,8 @@ Type * Compiler::CppToLlvmType() {
return m_ir_builder->getFloatTy();
} else if (std::is_same<T, double>::value) {
return m_ir_builder->getDoubleTy();
} else if (std::is_same<T, bool>::value) {
return m_ir_builder->getInt1Ty();
} else if (std::is_pointer<T>::value) {
return m_ir_builder->getInt8PtrTy();
} else {
@ -5664,7 +5551,17 @@ llvm::Value * Compiler::IndirectCall(u32 address, Value * context_i64, bool is_f
auto location_i64_ptr = m_ir_builder->CreateIntToPtr(location_i64, m_ir_builder->getInt64Ty()->getPointerTo());
auto executable_i64 = m_ir_builder->CreateLoad(location_i64_ptr);
auto executable_ptr = m_ir_builder->CreateIntToPtr(executable_i64, m_compiled_function_type->getPointerTo());
return m_ir_builder->CreateCall2(executable_ptr, m_state.args[CompileTaskState::Args::State], context_i64);
auto ret_i32 = m_ir_builder->CreateCall2(executable_ptr, m_state.args[CompileTaskState::Args::State], context_i64);
auto cmp_i1 = m_ir_builder->CreateICmpEQ(ret_i32, m_ir_builder->getInt32(0xFFFFFFFF));
auto then_bb = GetBasicBlockFromAddress(m_state.current_instruction_address, "then_all_fs");
auto merge_bb = GetBasicBlockFromAddress(m_state.current_instruction_address, "merge_all_fs");
m_ir_builder->CreateCondBr(cmp_i1, then_bb, merge_bb);
m_ir_builder->SetInsertPoint(then_bb);
m_ir_builder->CreateRet(m_ir_builder->getInt32(0));
m_ir_builder->SetInsertPoint(merge_bb);
return ret_i32;
}
void Compiler::CompilationError(const std::string & error) {
@ -5688,7 +5585,7 @@ RecompilationEngine::RecompilationEngine()
: ThreadBase("PPU Recompilation Engine")
, m_log(nullptr)
, m_next_ordinal(0)
, m_compiler(*this, ExecutionEngine::ExecuteFunction, ExecutionEngine::ExecuteTillReturn) {
, m_compiler(*this, ExecutionEngine::ExecuteFunction, ExecutionEngine::ExecuteTillReturn, ExecutionEngine::PollStatus) {
m_compiler.RunAllTests();
}
@ -6101,12 +5998,7 @@ u32 ppu_recompiler_llvm::ExecutionEngine::ExecuteTillReturn(PPUThread * ppu_stat
execution_engine->m_tracer.Trace(Tracer::TraceType::ExitFromCompiledFunction, context >> 32, context & 0xFFFFFFFF);
}
while (!terminate && !Emu.IsStopped()) {
if (Emu.IsPaused()) {
std::this_thread::sleep_for(std::chrono::milliseconds(50));
continue;
}
while (terminate == false && PollStatus(ppu_state) == false) {
auto executable = execution_engine->GetExecutable(ppu_state->PC, ExecuteTillReturn);
if (executable != ExecuteTillReturn && executable != ExecuteFunction) {
auto entry = ppu_state->PC;
@ -6146,6 +6038,18 @@ u32 ppu_recompiler_llvm::ExecutionEngine::ExecuteTillReturn(PPUThread * ppu_stat
return 0;
}
bool ppu_recompiler_llvm::ExecutionEngine::PollStatus(PPUThread * ppu_state) {
while (Emu.IsPaused() || ppu_state->IsPaused()) {
std::this_thread::sleep_for(std::chrono::milliseconds(50));
}
if (Emu.IsStopped() || ppu_state->IsStopped()) {
return true;
}
return false;
}
BranchType ppu_recompiler_llvm::GetBranchTypeFromInstruction(u32 instruction) {
auto type = BranchType::NonBranch;
auto field1 = instruction >> 26;

15
rpcs3/Emu/Cell/PPULLVMRecompiler.h
View File

@ -272,7 +272,8 @@ namespace ppu_recompiler_llvm {
std::chrono::nanoseconds total_time;
};
Compiler(RecompilationEngine & recompilation_engine, const Executable execute_unknown_function, const Executable execute_unknown_block);
Compiler(RecompilationEngine & recompilation_engine, const Executable execute_unknown_function,
const Executable execute_unknown_block, bool (*poll_status_function)(PPUThread * ppu_state));
Compiler(const Compiler & other) = delete;
Compiler(Compiler && other) = delete;
@ -469,7 +470,7 @@ namespace ppu_recompiler_llvm {
void ADDI(u32 rd, u32 ra, s32 simm16) override;
void ADDIS(u32 rd, u32 ra, s32 simm16) override;
void BC(u32 bo, u32 bi, s32 bd, u32 aa, u32 lk) override;
void HACK(u32 id) override;
void HACK(u32 id) override;
void SC(u32 sc_code) override;
void B(s32 ll, u32 aa, u32 lk) override;
void MCRF(u32 crfd, u32 crfs) override;
@ -735,6 +736,9 @@ namespace ppu_recompiler_llvm {
/// Recompilation engine
RecompilationEngine & m_recompilation_engine;
/// The function that should be called to check the status of the thread
bool (*m_poll_status_function)(PPUThread * ppu_state);
/// The function that will be called to execute unknown functions
llvm::Function * m_execute_unknown_function;
@ -924,8 +928,8 @@ namespace ppu_recompiler_llvm {
llvm::Value * IndirectCall(u32 address, llvm::Value * context_i64, bool is_function);
/// Test an instruction against the interpreter
template <class PPULLVMRecompilerFn, class PPUInterpreterFn, class... Args>
void VerifyInstructionAgainstInterpreter(const char * name, PPULLVMRecompilerFn recomp_fn, PPUInterpreterFn interp_fn, PPUState & input_state, Args... args);
template <class... Args>
void VerifyInstructionAgainstInterpreter(const char * name, void (Compiler::*recomp_fn)(Args...), void (PPUInterpreter::*interp_fn)(Args...), PPUState & input_state, Args... args);
/// Excute a test
void RunTest(const char * name, std::function<void()> test_case, std::function<void()> input, std::function<bool(std::string & msg)> check_result);
@ -1164,6 +1168,9 @@ namespace ppu_recompiler_llvm {
/// Execute till the current function returns
static u32 ExecuteTillReturn(PPUThread * ppu_state, u64 context);
/// Check thread status. Returns true if the thread must exit.
static bool PollStatus(PPUThread * ppu_state);
};
/// Get the branch type from a branch instruction

1082
rpcs3/Emu/Cell/PPULLVMRecompilerTests.cpp
View File

File diff suppressed because it is too large Load Diff

5
rpcs3/Emu/Memory/vm.cpp
View File

@ -242,6 +242,11 @@ namespace vm
return broken;
}
bool reservation_acquire_no_cb(void* data, u32 addr, u32 size)
{
return reservation_acquire(data, addr, size);
}
bool reservation_update(u32 addr, const void* data, u32 size)
{
assert(size == 1 || size == 2 || size == 4 || size == 8 || size == 128);

3
rpcs3/Emu/Memory/vm.h
View File

@ -36,6 +36,9 @@ namespace vm
bool reservation_break(u32 addr);
// read memory and reserve it for further atomic update, return true if the previous reservation was broken
bool reservation_acquire(void* data, u32 addr, u32 size, const std::function<void()>& callback = nullptr);
// same as reservation_acquire but does not have the callback argument
// used by the PPU LLVM JIT since creating a std::function object in LLVM IR is too complicated
bool reservation_acquire_no_cb(void* data, u32 addr, u32 size);
// attempt to atomically update reserved memory
bool reservation_update(u32 addr, const void* data, u32 size);
// for internal use