Disabled optimizations, block merging and instruction skipping code while the debugger is stepping.
This commit is contained in:
parent
4570dd7eeb
commit
daf977e84e
|
@ -190,7 +190,7 @@ void Jit64::Init()
|
||||||
|
|
||||||
// BLR optimization has the same consequences as block linking, as well as
|
// BLR optimization has the same consequences as block linking, as well as
|
||||||
// depending on the fault handler to be safe in the event of excessive BL.
|
// depending on the fault handler to be safe in the event of excessive BL.
|
||||||
m_enable_blr_optimization = jo.enableBlocklink && SConfig::GetInstance().m_LocalCoreStartupParameter.bFastmem;
|
m_enable_blr_optimization = jo.enableBlocklink && SConfig::GetInstance().m_LocalCoreStartupParameter.bFastmem && !SConfig::GetInstance().m_LocalCoreStartupParameter.bEnableDebugging;
|
||||||
m_clear_cache_asap = false;
|
m_clear_cache_asap = false;
|
||||||
|
|
||||||
m_stack = nullptr;
|
m_stack = nullptr;
|
||||||
|
@ -207,9 +207,7 @@ void Jit64::Init()
|
||||||
code_block.m_stats = &js.st;
|
code_block.m_stats = &js.st;
|
||||||
code_block.m_gpa = &js.gpa;
|
code_block.m_gpa = &js.gpa;
|
||||||
code_block.m_fpa = &js.fpa;
|
code_block.m_fpa = &js.fpa;
|
||||||
analyzer.SetOption(PPCAnalyst::PPCAnalyzer::OPTION_CONDITIONAL_CONTINUE);
|
EnableOptimization();
|
||||||
analyzer.SetOption(PPCAnalyst::PPCAnalyzer::OPTION_BRANCH_MERGE);
|
|
||||||
analyzer.SetOption(PPCAnalyst::PPCAnalyzer::OPTION_CARRY_MERGE);
|
|
||||||
}
|
}
|
||||||
|
|
||||||
void Jit64::ClearCache()
|
void Jit64::ClearCache()
|
||||||
|
@ -518,6 +516,7 @@ const u8* Jit64::DoJit(u32 em_address, PPCAnalyst::CodeBuffer *code_buf, JitBloc
|
||||||
{
|
{
|
||||||
// We can link blocks as long as we are not single stepping and there are no breakpoints here
|
// We can link blocks as long as we are not single stepping and there are no breakpoints here
|
||||||
EnableBlockLink();
|
EnableBlockLink();
|
||||||
|
EnableOptimization();
|
||||||
|
|
||||||
// Comment out the following to disable breakpoints (speed-up)
|
// Comment out the following to disable breakpoints (speed-up)
|
||||||
if (!Profiler::g_ProfileBlocks)
|
if (!Profiler::g_ProfileBlocks)
|
||||||
|
@ -528,6 +527,9 @@ const u8* Jit64::DoJit(u32 em_address, PPCAnalyst::CodeBuffer *code_buf, JitBloc
|
||||||
|
|
||||||
// Do not link this block to other blocks While single stepping
|
// Do not link this block to other blocks While single stepping
|
||||||
jo.enableBlocklink = false;
|
jo.enableBlocklink = false;
|
||||||
|
analyzer.ClearOption(PPCAnalyst::PPCAnalyzer::OPTION_CONDITIONAL_CONTINUE);
|
||||||
|
analyzer.ClearOption(PPCAnalyst::PPCAnalyzer::OPTION_BRANCH_MERGE);
|
||||||
|
analyzer.ClearOption(PPCAnalyst::PPCAnalyzer::OPTION_CARRY_MERGE);
|
||||||
}
|
}
|
||||||
Trace();
|
Trace();
|
||||||
}
|
}
|
||||||
|
@ -873,3 +875,10 @@ void Jit64::EnableBlockLink()
|
||||||
jo.enableBlocklink = false;
|
jo.enableBlocklink = false;
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
|
void Jit64::EnableOptimization()
|
||||||
|
{
|
||||||
|
analyzer.SetOption(PPCAnalyst::PPCAnalyzer::OPTION_CONDITIONAL_CONTINUE);
|
||||||
|
analyzer.SetOption(PPCAnalyst::PPCAnalyzer::OPTION_BRANCH_MERGE);
|
||||||
|
analyzer.SetOption(PPCAnalyst::PPCAnalyzer::OPTION_CARRY_MERGE);
|
||||||
|
}
|
||||||
|
|
|
@ -65,6 +65,8 @@ public:
|
||||||
|
|
||||||
void Init() override;
|
void Init() override;
|
||||||
|
|
||||||
|
void EnableOptimization();
|
||||||
|
|
||||||
void EnableBlockLink();
|
void EnableBlockLink();
|
||||||
|
|
||||||
void Shutdown() override;
|
void Shutdown() override;
|
||||||
|
|
|
@ -97,11 +97,14 @@ void Jit64::lXXx(UGeckoInstruction inst)
|
||||||
// PowerPC has no 8-bit sign extended load, but x86 does, so merge extsb with the load if we find it.
|
// PowerPC has no 8-bit sign extended load, but x86 does, so merge extsb with the load if we find it.
|
||||||
if (accessSize == 8 && js.next_inst.OPCD == 31 && js.next_inst.SUBOP10 == 954 &&
|
if (accessSize == 8 && js.next_inst.OPCD == 31 && js.next_inst.SUBOP10 == 954 &&
|
||||||
js.next_inst.RS == inst.RD && js.next_inst.RA == inst.RD && !js.next_inst.Rc)
|
js.next_inst.RS == inst.RD && js.next_inst.RA == inst.RD && !js.next_inst.Rc)
|
||||||
|
{
|
||||||
|
if (PowerPC::GetState() != PowerPC::CPU_STEPPING)
|
||||||
{
|
{
|
||||||
js.downcountAmount++;
|
js.downcountAmount++;
|
||||||
js.skipnext = true;
|
js.skipnext = true;
|
||||||
signExtend = true;
|
signExtend = true;
|
||||||
}
|
}
|
||||||
|
}
|
||||||
|
|
||||||
// TODO(ector): Make it dynamically enable/disable idle skipping where appropriate
|
// TODO(ector): Make it dynamically enable/disable idle skipping where appropriate
|
||||||
// Will give nice boost to dual core mode
|
// Will give nice boost to dual core mode
|
||||||
|
@ -109,6 +112,7 @@ void Jit64::lXXx(UGeckoInstruction inst)
|
||||||
// IMHO those Idles should always be skipped and replaced by a more controllable "native" Idle methode
|
// IMHO those Idles should always be skipped and replaced by a more controllable "native" Idle methode
|
||||||
// ... maybe the throttle one already do that :p
|
// ... maybe the throttle one already do that :p
|
||||||
if (SConfig::GetInstance().m_LocalCoreStartupParameter.bSkipIdle &&
|
if (SConfig::GetInstance().m_LocalCoreStartupParameter.bSkipIdle &&
|
||||||
|
PowerPC::GetState() != PowerPC::CPU_STEPPING &&
|
||||||
inst.OPCD == 32 &&
|
inst.OPCD == 32 &&
|
||||||
(inst.hex & 0xFFFF0000) == 0x800D0000 &&
|
(inst.hex & 0xFFFF0000) == 0x800D0000 &&
|
||||||
(Memory::ReadUnchecked_U32(js.compilerPC + 4) == 0x28000000 ||
|
(Memory::ReadUnchecked_U32(js.compilerPC + 4) == 0x28000000 ||
|
||||||
|
|
|
@ -227,6 +227,8 @@ void Jit64::mfspr(UGeckoInstruction inst)
|
||||||
u32 nextIndex = (js.next_inst.SPRU << 5) | (js.next_inst.SPRL & 0x1F);
|
u32 nextIndex = (js.next_inst.SPRU << 5) | (js.next_inst.SPRL & 0x1F);
|
||||||
// Be careful; the actual opcode is for mftb (371), not mfspr (339)
|
// Be careful; the actual opcode is for mftb (371), not mfspr (339)
|
||||||
if (js.next_inst.OPCD == 31 && js.next_inst.SUBOP10 == 371 && (nextIndex == SPR_TU || nextIndex == SPR_TL))
|
if (js.next_inst.OPCD == 31 && js.next_inst.SUBOP10 == 371 && (nextIndex == SPR_TU || nextIndex == SPR_TL))
|
||||||
|
{
|
||||||
|
if (PowerPC::GetState() != PowerPC::CPU_STEPPING)
|
||||||
{
|
{
|
||||||
int n = js.next_inst.RD;
|
int n = js.next_inst.RD;
|
||||||
js.downcountAmount++;
|
js.downcountAmount++;
|
||||||
|
@ -244,6 +246,7 @@ void Jit64::mfspr(UGeckoInstruction inst)
|
||||||
if (nextIndex == SPR_TU)
|
if (nextIndex == SPR_TU)
|
||||||
MOV(32, gpr.R(n), R(RAX));
|
MOV(32, gpr.R(n), R(RAX));
|
||||||
}
|
}
|
||||||
|
}
|
||||||
else
|
else
|
||||||
{
|
{
|
||||||
gpr.Lock(d);
|
gpr.Lock(d);
|
||||||
|
|
|
@ -37,6 +37,7 @@ void JitILBase::lXz(UGeckoInstruction inst)
|
||||||
// Idle Skipping. This really should be done somewhere else.
|
// Idle Skipping. This really should be done somewhere else.
|
||||||
// Either lower in the IR or higher in PPCAnalyist
|
// Either lower in the IR or higher in PPCAnalyist
|
||||||
if (SConfig::GetInstance().m_LocalCoreStartupParameter.bSkipIdle &&
|
if (SConfig::GetInstance().m_LocalCoreStartupParameter.bSkipIdle &&
|
||||||
|
PowerPC::GetState() != PowerPC::CPU_STEPPING &&
|
||||||
inst.OPCD == 32 && // Lwx
|
inst.OPCD == 32 && // Lwx
|
||||||
(inst.hex & 0xFFFF0000) == 0x800D0000 &&
|
(inst.hex & 0xFFFF0000) == 0x800D0000 &&
|
||||||
(Memory::ReadUnchecked_U32(js.compilerPC + 4) == 0x28000000 ||
|
(Memory::ReadUnchecked_U32(js.compilerPC + 4) == 0x28000000 ||
|
||||||
|
|
Loading…
Reference in New Issue