summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat (limited to '')
m---------externals/dynarmic0
-rw-r--r--src/core/arm/dynarmic/arm_dynarmic_32.cpp13
-rw-r--r--src/core/arm/dynarmic/arm_dynarmic_64.cpp13
-rw-r--r--src/core/core_timing.cpp12
-rw-r--r--src/core/core_timing_util.cpp1
-rw-r--r--src/core/core_timing_util.h1
-rw-r--r--src/core/settings.h6
-rw-r--r--src/video_core/macro/macro_interpreter.cpp1
-rw-r--r--src/video_core/renderer_opengl/gl_shader_disk_cache.cpp23
-rw-r--r--src/yuzu/configuration/config.cpp10
-rw-r--r--src/yuzu/configuration/configure_cpu.cpp17
-rw-r--r--src/yuzu/configuration/configure_cpu.h1
-rw-r--r--src/yuzu/configuration/configure_cpu.ui52
13 files changed, 126 insertions, 24 deletions
diff --git a/externals/dynarmic b/externals/dynarmic
-Subproject 82417da7803e2cf18efc28a1cd3f3d0a4b6045a
+Subproject 0e1112b7df77ae55a62a51622940d5c8f9e8c84
diff --git a/src/core/arm/dynarmic/arm_dynarmic_32.cpp b/src/core/arm/dynarmic/arm_dynarmic_32.cpp
index 443ca72eb..b5f28a86e 100644
--- a/src/core/arm/dynarmic/arm_dynarmic_32.cpp
+++ b/src/core/arm/dynarmic/arm_dynarmic_32.cpp
@@ -143,7 +143,7 @@ std::shared_ptr<Dynarmic::A32::Jit> ARM_Dynarmic_32::MakeJit(Common::PageTable&
config.wall_clock_cntpct = uses_wall_clock;
// Safe optimizations
- if (Settings::values.cpu_accuracy != Settings::CPUAccuracy::Accurate) {
+ if (Settings::values.cpu_accuracy == Settings::CPUAccuracy::DebugMode) {
if (!Settings::values.cpuopt_page_tables) {
config.page_table = nullptr;
}
@@ -170,6 +170,17 @@ std::shared_ptr<Dynarmic::A32::Jit> ARM_Dynarmic_32::MakeJit(Common::PageTable&
}
}
+ // Unsafe optimizations
+ if (Settings::values.cpu_accuracy == Settings::CPUAccuracy::Unsafe) {
+ config.unsafe_optimizations = true;
+ if (Settings::values.cpuopt_unsafe_unfuse_fma) {
+ config.optimizations |= Dynarmic::OptimizationFlag::Unsafe_UnfuseFMA;
+ }
+ if (Settings::values.cpuopt_unsafe_reduce_fp_error) {
+ config.optimizations |= Dynarmic::OptimizationFlag::Unsafe_ReducedErrorFP;
+ }
+ }
+
return std::make_unique<Dynarmic::A32::Jit>(config);
}
diff --git a/src/core/arm/dynarmic/arm_dynarmic_64.cpp b/src/core/arm/dynarmic/arm_dynarmic_64.cpp
index a63a04a25..ce9968724 100644
--- a/src/core/arm/dynarmic/arm_dynarmic_64.cpp
+++ b/src/core/arm/dynarmic/arm_dynarmic_64.cpp
@@ -195,7 +195,7 @@ std::shared_ptr<Dynarmic::A64::Jit> ARM_Dynarmic_64::MakeJit(Common::PageTable&
config.wall_clock_cntpct = uses_wall_clock;
// Safe optimizations
- if (Settings::values.cpu_accuracy != Settings::CPUAccuracy::Accurate) {
+ if (Settings::values.cpu_accuracy == Settings::CPUAccuracy::DebugMode) {
if (!Settings::values.cpuopt_page_tables) {
config.page_table = nullptr;
}
@@ -222,6 +222,17 @@ std::shared_ptr<Dynarmic::A64::Jit> ARM_Dynarmic_64::MakeJit(Common::PageTable&
}
}
+ // Unsafe optimizations
+ if (Settings::values.cpu_accuracy == Settings::CPUAccuracy::Unsafe) {
+ config.unsafe_optimizations = true;
+ if (Settings::values.cpuopt_unsafe_unfuse_fma) {
+ config.optimizations |= Dynarmic::OptimizationFlag::Unsafe_UnfuseFMA;
+ }
+ if (Settings::values.cpuopt_unsafe_reduce_fp_error) {
+ config.optimizations |= Dynarmic::OptimizationFlag::Unsafe_ReducedErrorFP;
+ }
+ }
+
return std::make_shared<Dynarmic::A64::Jit>(config);
}
diff --git a/src/core/core_timing.cpp b/src/core/core_timing.cpp
index 71af26ec5..e6c8461a5 100644
--- a/src/core/core_timing.cpp
+++ b/src/core/core_timing.cpp
@@ -7,14 +7,14 @@
#include <string>
#include <tuple>
-#include "common/assert.h"
#include "common/microprofile.h"
#include "core/core_timing.h"
#include "core/core_timing_util.h"
+#include "core/hardware_properties.h"
namespace Core::Timing {
-constexpr u64 MAX_SLICE_LENGTH = 4000;
+constexpr s64 MAX_SLICE_LENGTH = 4000;
std::shared_ptr<EventType> CreateEvent(std::string name, TimedCallback&& callback) {
return std::make_shared<EventType>(std::move(callback), std::move(name));
@@ -37,10 +37,8 @@ struct CoreTiming::Event {
}
};
-CoreTiming::CoreTiming() {
- clock =
- Common::CreateBestMatchingClock(Core::Hardware::BASE_CLOCK_RATE, Core::Hardware::CNTFREQ);
-}
+CoreTiming::CoreTiming()
+ : clock{Common::CreateBestMatchingClock(Hardware::BASE_CLOCK_RATE, Hardware::CNTFREQ)} {}
CoreTiming::~CoreTiming() = default;
@@ -136,7 +134,7 @@ void CoreTiming::UnscheduleEvent(const std::shared_ptr<EventType>& event_type,
void CoreTiming::AddTicks(u64 ticks) {
this->ticks += ticks;
- downcount -= ticks;
+ downcount -= static_cast<s64>(ticks);
}
void CoreTiming::Idle() {
diff --git a/src/core/core_timing_util.cpp b/src/core/core_timing_util.cpp
index aefc63663..8ce8e602e 100644
--- a/src/core/core_timing_util.cpp
+++ b/src/core/core_timing_util.cpp
@@ -8,6 +8,7 @@
#include <limits>
#include "common/logging/log.h"
#include "common/uint128.h"
+#include "core/hardware_properties.h"
namespace Core::Timing {
diff --git a/src/core/core_timing_util.h b/src/core/core_timing_util.h
index 2ed979e14..e4a046bf9 100644
--- a/src/core/core_timing_util.h
+++ b/src/core/core_timing_util.h
@@ -6,7 +6,6 @@
#include <chrono>
#include "common/common_types.h"
-#include "core/hardware_properties.h"
namespace Core::Timing {
diff --git a/src/core/settings.h b/src/core/settings.h
index bb145f193..3681b5e9d 100644
--- a/src/core/settings.h
+++ b/src/core/settings.h
@@ -359,7 +359,8 @@ enum class GPUAccuracy : u32 {
enum class CPUAccuracy {
Accurate = 0,
- DebugMode = 1,
+ Unsafe = 1,
+ DebugMode = 2,
};
extern bool configuring_global;
@@ -419,6 +420,9 @@ struct Values {
bool cpuopt_misc_ir;
bool cpuopt_reduce_misalign_checks;
+ bool cpuopt_unsafe_unfuse_fma;
+ bool cpuopt_unsafe_reduce_fp_error;
+
// Renderer
Setting<RendererBackend> renderer_backend;
bool renderer_debug;
diff --git a/src/video_core/macro/macro_interpreter.cpp b/src/video_core/macro/macro_interpreter.cpp
index aa5256419..bd01fd1f2 100644
--- a/src/video_core/macro/macro_interpreter.cpp
+++ b/src/video_core/macro/macro_interpreter.cpp
@@ -34,7 +34,6 @@ void MacroInterpreterImpl::Execute(const std::vector<u32>& parameters, u32 metho
this->parameters = std::make_unique<u32[]>(num_parameters);
}
std::memcpy(this->parameters.get(), parameters.data(), num_parameters * sizeof(u32));
- this->num_parameters = num_parameters;
// Execute the code until we hit an exit condition.
bool keep_executing = true;
diff --git a/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp b/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp
index 52fbab3c1..40c0877c1 100644
--- a/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_disk_cache.cpp
@@ -214,20 +214,20 @@ std::optional<std::vector<ShaderDiskCacheEntry>> ShaderDiskCacheOpenGL::LoadTran
// Skip games without title id
const bool has_title_id = system.CurrentProcess()->GetTitleID() != 0;
if (!Settings::values.use_disk_shader_cache.GetValue() || !has_title_id) {
- return {};
+ return std::nullopt;
}
Common::FS::IOFile file(GetTransferablePath(), "rb");
if (!file.IsOpen()) {
LOG_INFO(Render_OpenGL, "No transferable shader cache found");
is_usable = true;
- return {};
+ return std::nullopt;
}
u32 version{};
if (file.ReadBytes(&version, sizeof(version)) != sizeof(version)) {
LOG_ERROR(Render_OpenGL, "Failed to get transferable cache version, skipping it");
- return {};
+ return std::nullopt;
}
if (version < NativeVersion) {
@@ -235,12 +235,12 @@ std::optional<std::vector<ShaderDiskCacheEntry>> ShaderDiskCacheOpenGL::LoadTran
file.Close();
InvalidateTransferable();
is_usable = true;
- return {};
+ return std::nullopt;
}
if (version > NativeVersion) {
LOG_WARNING(Render_OpenGL, "Transferable shader cache was generated with a newer version "
"of the emulator, skipping");
- return {};
+ return std::nullopt;
}
// Version is valid, load the shaders
@@ -249,7 +249,7 @@ std::optional<std::vector<ShaderDiskCacheEntry>> ShaderDiskCacheOpenGL::LoadTran
ShaderDiskCacheEntry& entry = entries.emplace_back();
if (!entry.Load(file)) {
LOG_ERROR(Render_OpenGL, "Failed to load transferable raw entry, skipping");
- return {};
+ return std::nullopt;
}
}
@@ -290,12 +290,12 @@ std::optional<std::vector<ShaderDiskCachePrecompiled>> ShaderDiskCacheOpenGL::Lo
ShaderCacheVersionHash file_hash{};
if (!LoadArrayFromPrecompiled(file_hash.data(), file_hash.size())) {
precompiled_cache_virtual_file_offset = 0;
- return {};
+ return std::nullopt;
}
if (GetShaderCacheVersionHash() != file_hash) {
LOG_INFO(Render_OpenGL, "Precompiled cache is from another version of the emulator");
precompiled_cache_virtual_file_offset = 0;
- return {};
+ return std::nullopt;
}
std::vector<ShaderDiskCachePrecompiled> entries;
@@ -305,15 +305,16 @@ std::optional<std::vector<ShaderDiskCachePrecompiled>> ShaderDiskCacheOpenGL::Lo
if (!LoadObjectFromPrecompiled(entry.unique_identifier) ||
!LoadObjectFromPrecompiled(entry.binary_format) ||
!LoadObjectFromPrecompiled(binary_size)) {
- return {};
+ return std::nullopt;
}
entry.binary.resize(binary_size);
if (!LoadArrayFromPrecompiled(entry.binary.data(), entry.binary.size())) {
- return {};
+ return std::nullopt;
}
}
- return entries;
+
+ return std::move(entries);
}
void ShaderDiskCacheOpenGL::InvalidateTransferable() {
diff --git a/src/yuzu/configuration/config.cpp b/src/yuzu/configuration/config.cpp
index a372190cc..7af974d8d 100644
--- a/src/yuzu/configuration/config.cpp
+++ b/src/yuzu/configuration/config.cpp
@@ -635,6 +635,11 @@ void Config::ReadCpuValues() {
ReadSetting(QStringLiteral("cpuopt_misc_ir"), true).toBool();
Settings::values.cpuopt_reduce_misalign_checks =
ReadSetting(QStringLiteral("cpuopt_reduce_misalign_checks"), true).toBool();
+
+ Settings::values.cpuopt_unsafe_unfuse_fma =
+ ReadSetting(QStringLiteral("cpuopt_unsafe_unfuse_fma"), true).toBool();
+ Settings::values.cpuopt_unsafe_reduce_fp_error =
+ ReadSetting(QStringLiteral("cpuopt_unsafe_reduce_fp_error"), true).toBool();
}
qt_config->endGroup();
@@ -1132,6 +1137,11 @@ void Config::SaveCpuValues() {
WriteSetting(QStringLiteral("cpuopt_misc_ir"), Settings::values.cpuopt_misc_ir, true);
WriteSetting(QStringLiteral("cpuopt_reduce_misalign_checks"),
Settings::values.cpuopt_reduce_misalign_checks, true);
+
+ WriteSetting(QStringLiteral("cpuopt_unsafe_unfuse_fma"),
+ Settings::values.cpuopt_unsafe_unfuse_fma, true);
+ WriteSetting(QStringLiteral("cpuopt_unsafe_reduce_fp_error"),
+ Settings::values.cpuopt_unsafe_reduce_fp_error, true);
}
qt_config->endGroup();
diff --git a/src/yuzu/configuration/configure_cpu.cpp b/src/yuzu/configuration/configure_cpu.cpp
index 7493e5ffb..37fcd6adc 100644
--- a/src/yuzu/configuration/configure_cpu.cpp
+++ b/src/yuzu/configuration/configure_cpu.cpp
@@ -19,6 +19,8 @@ ConfigureCpu::ConfigureCpu(QWidget* parent) : QWidget(parent), ui(new Ui::Config
connect(ui->accuracy, qOverload<int>(&QComboBox::activated), this,
&ConfigureCpu::AccuracyUpdated);
+ connect(ui->accuracy, qOverload<int>(&QComboBox::currentIndexChanged), this,
+ &ConfigureCpu::UpdateGroup);
}
ConfigureCpu::~ConfigureCpu() = default;
@@ -28,6 +30,12 @@ void ConfigureCpu::SetConfiguration() {
ui->accuracy->setEnabled(runtime_lock);
ui->accuracy->setCurrentIndex(static_cast<int>(Settings::values.cpu_accuracy));
+ UpdateGroup(static_cast<int>(Settings::values.cpu_accuracy));
+
+ ui->cpuopt_unsafe_unfuse_fma->setEnabled(runtime_lock);
+ ui->cpuopt_unsafe_unfuse_fma->setChecked(Settings::values.cpuopt_unsafe_unfuse_fma);
+ ui->cpuopt_unsafe_reduce_fp_error->setEnabled(runtime_lock);
+ ui->cpuopt_unsafe_reduce_fp_error->setChecked(Settings::values.cpuopt_unsafe_reduce_fp_error);
}
void ConfigureCpu::AccuracyUpdated(int index) {
@@ -38,14 +46,21 @@ void ConfigureCpu::AccuracyUpdated(int index) {
QMessageBox::Yes | QMessageBox::No);
if (result == QMessageBox::No) {
ui->accuracy->setCurrentIndex(static_cast<int>(Settings::CPUAccuracy::Accurate));
- return;
+ UpdateGroup(static_cast<int>(Settings::CPUAccuracy::Accurate));
}
}
}
+void ConfigureCpu::UpdateGroup(int index) {
+ ui->unsafe_group->setVisible(static_cast<Settings::CPUAccuracy>(index) ==
+ Settings::CPUAccuracy::Unsafe);
+}
+
void ConfigureCpu::ApplyConfiguration() {
Settings::values.cpu_accuracy =
static_cast<Settings::CPUAccuracy>(ui->accuracy->currentIndex());
+ Settings::values.cpuopt_unsafe_unfuse_fma = ui->cpuopt_unsafe_unfuse_fma->isChecked();
+ Settings::values.cpuopt_unsafe_reduce_fp_error = ui->cpuopt_unsafe_reduce_fp_error->isChecked();
}
void ConfigureCpu::changeEvent(QEvent* event) {
diff --git a/src/yuzu/configuration/configure_cpu.h b/src/yuzu/configuration/configure_cpu.h
index e4741d3a4..3c5683d81 100644
--- a/src/yuzu/configuration/configure_cpu.h
+++ b/src/yuzu/configuration/configure_cpu.h
@@ -26,6 +26,7 @@ private:
void RetranslateUI();
void AccuracyUpdated(int index);
+ void UpdateGroup(int index);
void SetConfiguration();
diff --git a/src/yuzu/configuration/configure_cpu.ui b/src/yuzu/configuration/configure_cpu.ui
index bf6ea79bb..ebdd2e6e9 100644
--- a/src/yuzu/configuration/configure_cpu.ui
+++ b/src/yuzu/configuration/configure_cpu.ui
@@ -40,6 +40,11 @@
</item>
<item>
<property name="text">
+ <string>Unsafe</string>
+ </property>
+ </item>
+ <item>
+ <property name="text">
<string>Enable Debug Mode</string>
</property>
</item>
@@ -63,6 +68,53 @@
</layout>
</item>
<item>
+ <layout class="QVBoxLayout">
+ <item>
+ <widget class="QGroupBox" name="unsafe_group">
+ <property name="title">
+ <string>Unsafe CPU Optimization Settings</string>
+ </property>
+ <layout class="QVBoxLayout">
+ <item>
+ <widget class="QLabel">
+ <property name="wordWrap">
+ <bool>1</bool>
+ </property>
+ <property name="text">
+ <string>These settings reduce accuracy for speed.</string>
+ </property>
+ </widget>
+ </item>
+ <item>
+ <widget class="QCheckBox" name="cpuopt_unsafe_unfuse_fma">
+ <property name="text">
+ <string>Unfuse FMA (improve performance on CPUs without FMA)</string>
+ </property>
+ <property name="toolTip">
+ <string>
+ &lt;div&gt;This option improves speed by reducing accuracy of fused-multiply-add instructions on CPUs without native FMA support.&lt;/div&gt;
+ </string>
+ </property>
+ </widget>
+ </item>
+ <item>
+ <widget class="QCheckBox" name="cpuopt_unsafe_reduce_fp_error">
+ <property name="text">
+ <string>Faster FRSQRTE and FRECPE</string>
+ </property>
+ <property name="toolTip">
+ <string>
+ &lt;div&gt;This option improves the speed of some approximate floating-point functions by using less accurate native approximations.&lt;/div&gt;
+ </string>
+ </property>
+ </widget>
+ </item>
+ </layout>
+ </widget>
+ </item>
+ </layout>
+ </item>
+ <item>
<spacer name="verticalSpacer">
<property name="orientation">
<enum>Qt::Vertical</enum>
generated by cgit v1.2.3 (git 2.25.1) at 2024-08-15 05:27:39 +0200