dynarmic: Inline exclusive memory accesses

Inlines implementation of exclusive instructions into JITted code,
improving performance of applications relying heavily on these
instructions.

We also fastmem these instructions for additional speed, with
support for appropriate recompilation on fastmem failure.

An unsafe optimization to disable the intercore global_monitor is also
provided, should one wish to rely solely on cmpxchg semantics for
safety.

See also: merryhime/dynarmic#664
This commit is contained in:
merry
2022-02-27 19:40:05 +00:00
parent 96d90be59f
commit 16784e5bb3
17 changed files with 114 additions and 8 deletions

View File

@ -609,6 +609,7 @@ void Config::ReadCpuValues() {
ReadGlobalSetting(Settings::values.cpuopt_unsafe_ignore_standard_fpcr);
ReadGlobalSetting(Settings::values.cpuopt_unsafe_inaccurate_nan);
ReadGlobalSetting(Settings::values.cpuopt_unsafe_fastmem_check);
ReadGlobalSetting(Settings::values.cpuopt_unsafe_ignore_global_monitor);
if (global) {
ReadBasicSetting(Settings::values.cpu_debug_mode);
@ -621,6 +622,8 @@ void Config::ReadCpuValues() {
ReadBasicSetting(Settings::values.cpuopt_misc_ir);
ReadBasicSetting(Settings::values.cpuopt_reduce_misalign_checks);
ReadBasicSetting(Settings::values.cpuopt_fastmem);
ReadBasicSetting(Settings::values.cpuopt_fastmem_exclusives);
ReadBasicSetting(Settings::values.cpuopt_recompile_exclusives);
}
qt_config->endGroup();
@ -1139,6 +1142,7 @@ void Config::SaveCpuValues() {
WriteGlobalSetting(Settings::values.cpuopt_unsafe_ignore_standard_fpcr);
WriteGlobalSetting(Settings::values.cpuopt_unsafe_inaccurate_nan);
WriteGlobalSetting(Settings::values.cpuopt_unsafe_fastmem_check);
WriteGlobalSetting(Settings::values.cpuopt_unsafe_ignore_global_monitor);
if (global) {
WriteBasicSetting(Settings::values.cpu_debug_mode);

View File

@ -36,6 +36,7 @@ void ConfigureCpu::SetConfiguration() {
ui->cpuopt_unsafe_ignore_standard_fpcr->setEnabled(runtime_lock);
ui->cpuopt_unsafe_inaccurate_nan->setEnabled(runtime_lock);
ui->cpuopt_unsafe_fastmem_check->setEnabled(runtime_lock);
ui->cpuopt_unsafe_ignore_global_monitor->setEnabled(runtime_lock);
ui->cpuopt_unsafe_unfuse_fma->setChecked(Settings::values.cpuopt_unsafe_unfuse_fma.GetValue());
ui->cpuopt_unsafe_reduce_fp_error->setChecked(
@ -46,6 +47,8 @@ void ConfigureCpu::SetConfiguration() {
Settings::values.cpuopt_unsafe_inaccurate_nan.GetValue());
ui->cpuopt_unsafe_fastmem_check->setChecked(
Settings::values.cpuopt_unsafe_fastmem_check.GetValue());
ui->cpuopt_unsafe_ignore_global_monitor->setChecked(
Settings::values.cpuopt_unsafe_ignore_global_monitor.GetValue());
if (Settings::IsConfiguringGlobal()) {
ui->accuracy->setCurrentIndex(static_cast<int>(Settings::values.cpu_accuracy.GetValue()));
@ -82,6 +85,9 @@ void ConfigureCpu::ApplyConfiguration() {
ConfigurationShared::ApplyPerGameSetting(&Settings::values.cpuopt_unsafe_fastmem_check,
ui->cpuopt_unsafe_fastmem_check,
cpuopt_unsafe_fastmem_check);
ConfigurationShared::ApplyPerGameSetting(&Settings::values.cpuopt_unsafe_ignore_global_monitor,
ui->cpuopt_unsafe_ignore_global_monitor,
cpuopt_unsafe_ignore_global_monitor);
}
void ConfigureCpu::changeEvent(QEvent* event) {
@ -120,4 +126,7 @@ void ConfigureCpu::SetupPerGameUI() {
ConfigurationShared::SetColoredTristate(ui->cpuopt_unsafe_fastmem_check,
Settings::values.cpuopt_unsafe_fastmem_check,
cpuopt_unsafe_fastmem_check);
ConfigurationShared::SetColoredTristate(ui->cpuopt_unsafe_ignore_global_monitor,
Settings::values.cpuopt_unsafe_ignore_global_monitor,
cpuopt_unsafe_ignore_global_monitor);
}

View File

@ -45,6 +45,7 @@ private:
ConfigurationShared::CheckState cpuopt_unsafe_ignore_standard_fpcr;
ConfigurationShared::CheckState cpuopt_unsafe_inaccurate_nan;
ConfigurationShared::CheckState cpuopt_unsafe_fastmem_check;
ConfigurationShared::CheckState cpuopt_unsafe_ignore_global_monitor;
const Core::System& system;
};

View File

@ -150,6 +150,18 @@
</property>
</widget>
</item>
<item>
<widget class="QCheckBox" name="cpuopt_unsafe_ignore_global_monitor">
<property name="toolTip">
<string>
&lt;div&gt;This option improves speed by relying only on the semantics of cmpxchg to ensure safety of exclusive access instructions. Please note this may result in deadlocks and other race conditions.&lt;/div&gt;
</string>
</property>
<property name="text">
<string>Ignore global monitor</string>
</property>
</widget>
</item>
</layout>
</widget>
</item>

View File

@ -44,6 +44,12 @@ void ConfigureCpuDebug::SetConfiguration() {
Settings::values.cpuopt_reduce_misalign_checks.GetValue());
ui->cpuopt_fastmem->setEnabled(runtime_lock);
ui->cpuopt_fastmem->setChecked(Settings::values.cpuopt_fastmem.GetValue());
ui->cpuopt_fastmem_exclusives->setEnabled(runtime_lock);
ui->cpuopt_fastmem_exclusives->setChecked(
Settings::values.cpuopt_fastmem_exclusives.GetValue());
ui->cpuopt_recompile_exclusives->setEnabled(runtime_lock);
ui->cpuopt_recompile_exclusives->setChecked(
Settings::values.cpuopt_recompile_exclusives.GetValue());
}
void ConfigureCpuDebug::ApplyConfiguration() {
@ -56,6 +62,8 @@ void ConfigureCpuDebug::ApplyConfiguration() {
Settings::values.cpuopt_misc_ir = ui->cpuopt_misc_ir->isChecked();
Settings::values.cpuopt_reduce_misalign_checks = ui->cpuopt_reduce_misalign_checks->isChecked();
Settings::values.cpuopt_fastmem = ui->cpuopt_fastmem->isChecked();
Settings::values.cpuopt_fastmem_exclusives = ui->cpuopt_fastmem_exclusives->isChecked();
Settings::values.cpuopt_recompile_exclusives = ui->cpuopt_recompile_exclusives->isChecked();
}
void ConfigureCpuDebug::changeEvent(QEvent* event) {

View File

@ -144,7 +144,34 @@
</string>
</property>
<property name="text">
<string>Enable Host MMU Emulation</string>
<string>Enable Host MMU Emulation (general memory instructions)</string>
</property>
</widget>
</item>
<item>
<widget class="QCheckBox" name="cpuopt_fastmem_exclusives">
<property name="toolTip">
<string>
&lt;div style=&quot;white-space: nowrap&quot;&gt;This optimization speeds up exclusive memory accesses by the guest program.&lt;/div&gt;
&lt;div style=&quot;white-space: nowrap&quot;&gt;Enabling it causes guest exclusive memory reads/writes to be done directly into memory and make use of Host's MMU.&lt;/div&gt;
&lt;div style=&quot;white-space: nowrap&quot;&gt;Disabling this forces all exclusive memory accesses to use Software MMU Emulation.&lt;/div&gt;
</string>
</property>
<property name="text">
<string>Enable Host MMU Emulation (exclusive memory instructions)</string>
</property>
</widget>
</item>
<item>
<widget class="QCheckBox" name="cpuopt_recompile_exclusives">
<property name="toolTip">
<string>
&lt;div style=&quot;white-space: nowrap&quot;&gt;This optimization speeds up exclusive memory accesses by the guest program.&lt;/div&gt;
&lt;div style=&quot;white-space: nowrap&quot;&gt;Enabling it reduces the overhead of fastmem failure of exclusive memory accesses.&lt;/div&gt;
</string>
</property>
<property name="text">
<string>Enable recompilation of exclusive memory instructions</string>
</property>
</widget>
</item>