Compare commits

...

4 Commits

Author SHA1 Message Date
CamilleLaVey 9a8a30ba43 First intent on motion resolution flow auto 2026-08-13 21:17:09 -04:00
CamilleLaVey 59384de777 Revert the Auto policy 2026-08-13 20:42:49 -04:00
CamilleLaVey 9b20c6c609 Queues and refresh rate adjustments 2026-08-13 17:26:52 -04:00
CamilleLaVey 1428abe1d1 Ducky DUUUUUUUUUUUUUUCKS 2026-08-13 16:50:32 -04:00
14 changed files with 176 additions and 36 deletions
@@ -39,6 +39,7 @@ enum class BooleanSetting(override val key: String) : AbstractBooleanSetting {
RENDERER_SAMPLE_SHADING("sample_shading"),
RENDERER_FRAME_GEN("frame_gen"),
RENDERER_FRAME_GEN_FP16("frame_gen_fp16"),
RENDERER_FRAME_GEN_FLOW_SCALE_AUTO("frame_gen_flow_scale_auto"),
RENDERER_FRAME_GEN_DUMP_FLOW("frame_gen_dump_flow"),
GPU_UNSWIZZLE_ENABLED("gpu_unswizzle_enabled"),
PICTURE_IN_PICTURE("picture_in_picture"),
@@ -20,6 +20,7 @@ enum class IntSetting(override val key: String) : AbstractIntSetting {
RENDERER_ACCURACY("gpu_accuracy"),
RENDERER_RESOLUTION("resolution_setup"),
RENDERER_FRAME_GEN_MULTIPLIER("frame_gen_multiplier"),
RENDERER_FRAME_GEN_QUEUE_TARGET("frame_gen_queue_target"),
RENDERER_FRAME_GEN_FLOW_SCALE("frame_gen_flow_scale"),
RENDERER_VSYNC("use_vsync"),
RENDERER_SCALING_FILTER("scaling_filter"),
@@ -99,6 +99,8 @@ abstract class SettingsItem(
private val frameGenKeys = setOf(
BooleanSetting.RENDERER_FRAME_GEN.key,
IntSetting.RENDERER_FRAME_GEN_MULTIPLIER.key,
IntSetting.RENDERER_FRAME_GEN_QUEUE_TARGET.key,
BooleanSetting.RENDERER_FRAME_GEN_FLOW_SCALE_AUTO.key,
IntSetting.RENDERER_FRAME_GEN_FLOW_SCALE.key,
BooleanSetting.RENDERER_FRAME_GEN_FP16.key,
BooleanSetting.RENDERER_FRAME_GEN_DUMP_FLOW.key
@@ -638,6 +640,22 @@ abstract class SettingsItem(
valuesId = R.array.frameGenMultiplierValues
)
)
put(
SingleChoiceSetting(
IntSetting.RENDERER_FRAME_GEN_QUEUE_TARGET,
titleId = R.string.frame_gen_queue_target,
descriptionId = R.string.frame_gen_queue_target_description,
choicesId = R.array.frameGenQueueTargetNames,
valuesId = R.array.frameGenQueueTargetValues
)
)
put(
SwitchSetting(
BooleanSetting.RENDERER_FRAME_GEN_FLOW_SCALE_AUTO,
titleId = R.string.frame_gen_flow_scale_auto,
descriptionId = R.string.frame_gen_flow_scale_auto_description
)
)
put(
SliderSetting(
IntSetting.RENDERER_FRAME_GEN_FLOW_SCALE,
@@ -126,7 +126,14 @@ class SettingsFragmentPresenter(
add(BooleanSetting.RENDERER_FRAME_GEN.key)
add(IntSetting.RENDERER_FRAME_GEN_MULTIPLIER.key)
add(IntSetting.RENDERER_FRAME_GEN_FLOW_SCALE.key)
add(IntSetting.RENDERER_FRAME_GEN_QUEUE_TARGET.key)
add(BooleanSetting.RENDERER_FRAME_GEN_FLOW_SCALE_AUTO.key)
if (!BooleanSetting.RENDERER_FRAME_GEN_FLOW_SCALE_AUTO.getBoolean(
getNeedsGlobalForKey(BooleanSetting.RENDERER_FRAME_GEN_FLOW_SCALE_AUTO.key)
)
) {
add(IntSetting.RENDERER_FRAME_GEN_FLOW_SCALE.key)
}
add(BooleanSetting.RENDERER_FRAME_GEN_FP16.key)
}
}
@@ -221,6 +228,14 @@ class SettingsFragmentPresenter(
menuKey = MenuTag.SECTION_RENDERER
)
)
add(
SubmenuSetting(
titleId = R.string.frame_gen,
descriptionId = R.string.frame_gen_submenu_description,
iconId = R.drawable.ic_duck,
menuKey = MenuTag.SECTION_FRAME_GEN
)
)
if (!NativeConfig.isPerGameConfigLoaded()) {
add(
SubmenuSetting(
@@ -336,15 +351,6 @@ class SettingsFragmentPresenter(
}
add(IntSetting.RENDERER_ANTI_ALIASING.key)
add(
SubmenuSetting(
titleId = R.string.frame_gen,
descriptionId = R.string.frame_gen_submenu_description,
iconId = R.drawable.ic_frames,
menuKey = MenuTag.SECTION_FRAME_GEN
)
)
add(HeaderSetting(R.string.advanced))
add(IntSetting.RENDERER_ACCURACY.key)
@@ -0,0 +1,19 @@
<vector xmlns:android="http://schemas.android.com/apk/res/android"
android:width="24dp"
android:height="24dp"
android:viewportWidth="24"
android:viewportHeight="24">
<path
android:fillColor="?attr/colorControlNormal"
android:pathData="M11,10.6a8,5.2 0 1,0 0,10.4a8,5.2 0 1,0 0,-10.4z" />
<path
android:fillColor="?attr/colorControlNormal"
android:pathData="M4.6,12.6L0.8,10.2L3.2,15.5z" />
<path
android:fillColor="?attr/colorControlNormal"
android:fillType="evenOdd"
android:pathData="M15.3,3.5a4,4 0 1,0 0,8a4,4 0 1,0 0,-8zM16.6,4.9a1,1 0 1,0 0,2a1,1 0 1,0 0,-2z" />
<path
android:fillColor="?attr/colorControlNormal"
android:pathData="M18.6,6.5L23.2,7.8L18.6,9.3z" />
</vector>
@@ -174,6 +174,18 @@
<item>4</item>
</integer-array>
<string-array name="frameGenQueueTargetNames">
<item>@string/frame_gen_queue_target_0</item>
<item>@string/frame_gen_queue_target_1</item>
<item>@string/frame_gen_queue_target_2</item>
</string-array>
<integer-array name="frameGenQueueTargetValues">
<item>0</item>
<item>1</item>
<item>2</item>
</integer-array>
<string-array name="rendererVSyncNames">
<item>@string/renderer_vsync_immediate</item>
<item>@string/renderer_vsync_mailbox</item>
@@ -302,10 +302,17 @@
<string name="frame_gen_submenu_description">Manage and configure frame generation</string>
<string name="frame_gen_description">Insert interpolated frames between rendered ones using Lossless Scaling. Forces FIFO presentation while enabled.</string>
<string name="frame_gen_multiplier">Frame multiplier</string>
<string name="frame_gen_multiplier_description">How many frames to display for each rendered frame. Higher values cost proportionally more GPU time.</string>
<string name="frame_gen_multiplier_description">How many frames to display for each rendered frame. Higher values cost proportionally more GPU time. Asking for more than your display can present will slow emulation down.</string>
<string name="frame_gen_multiplier_2x">2x</string>
<string name="frame_gen_multiplier_3x">3x</string>
<string name="frame_gen_multiplier_4x">4x</string>
<string name="frame_gen_queue_target">Frame queue target</string>
<string name="frame_gen_queue_target_description">How many finished frames may wait ahead of the display. Larger queues absorb GPU spikes at the cost of input latency.</string>
<string name="frame_gen_queue_target_0">Unbuffered (lowest latency)</string>
<string name="frame_gen_queue_target_1">One frame (balanced)</string>
<string name="frame_gen_queue_target_2">Two frames (smoothest)</string>
<string name="frame_gen_flow_scale_auto">Match motion estimation to the game</string>
<string name="frame_gen_flow_scale_auto_description">Estimate motion at the resolution the game actually renders instead of the upscaled output. Costs nothing in accuracy, since upscaling adds no motion detail.</string>
<string name="frame_gen_flow_scale">Motion estimation resolution</string>
<string name="frame_gen_flow_scale_description">Resolution of the optical flow pass, as a fraction of the output. Lowering it is the cheapest way to reclaim performance.</string>
<string name="frame_gen_fp16">Half precision shaders</string>
+12
View File
@@ -380,6 +380,18 @@ void UpdateRescalingInfo() {
TranslateResolutionInfo(setup, info);
}
u32 FrameGenMultiplier() {
return std::clamp(values.frame_gen_multiplier.GetValue(), MIN_FRAME_GEN_MULTIPLIER,
MAX_FRAME_GEN_MULTIPLIER);
}
size_t FrameGenGenerations() {
if (!values.frame_gen.GetValue()) {
return 0;
}
return FrameGenMultiplier() - 1;
}
void RestoreGlobalState(bool is_powered_on) {
// If a game is running, DO NOT restore the global settings state
if (is_powered_on) {
+28
View File
@@ -402,6 +402,15 @@ struct Values {
false,
&frame_gen};
SwitchableSetting<bool> frame_gen_flow_scale_auto{linkage,
true,
"frame_gen_flow_scale_auto",
Category::Renderer,
Specialization::Default,
true,
false,
&frame_gen};
SwitchableSetting<u32, true> frame_gen_flow_scale{linkage,
75,
25,
@@ -414,6 +423,17 @@ struct Values {
true,
&frame_gen};
SwitchableSetting<u32, true> frame_gen_queue_target{linkage,
1,
0,
2,
"frame_gen_queue_target",
Category::Renderer,
Specialization::Countable,
true,
false,
&frame_gen};
SwitchableSetting<bool> frame_gen_fp16{linkage, true, "frame_gen_fp16", Category::Renderer,
Specialization::Default, true, false, &frame_gen};
@@ -902,10 +922,18 @@ struct Values {
// Per-game overrides
bool use_squashed_iterated_blend;
};
extern Values values;
constexpr u32 MIN_FRAME_GEN_MULTIPLIER = 2;
constexpr u32 MAX_FRAME_GEN_MULTIPLIER = 4;
[[nodiscard]] u32 FrameGenMultiplier();
[[nodiscard]] size_t FrameGenGenerations();
bool getDebugKnobAt(u8 i);
void UpdateGPUAccuracy();
@@ -22,14 +22,29 @@ namespace {
constexpr size_t COLOR_CHANNELS = 4;
constexpr u64 LSFG_REQUIRED_FRAMES = 2;
[[nodiscard]] f32 ConfiguredFlowScale() {
[[nodiscard]] f32 ManualFlowScale() {
return static_cast<f32>(Settings::values.frame_gen_flow_scale.GetValue()) / 100.0f;
}
[[nodiscard]] f32 ConfiguredFlowScale(VkExtent2D guest_extent, VkExtent2D presented_extent) {
if (!Settings::values.frame_gen_flow_scale_auto.GetValue()) {
return ManualFlowScale();
}
if (guest_extent.width == 0 || presented_extent.width == 0) {
return 1.0f;
}
const f32 rendered_width = static_cast<f32>(guest_extent.width) *
Settings::values.resolution_info.up_factor;
const f32 ratio = rendered_width / static_cast<f32>(presented_extent.width);
constexpr f32 FLOW_SCALE_STEPS = 20.0f;
const f32 stepped = std::ceil(ratio * FLOW_SCALE_STEPS) / FLOW_SCALE_STEPS;
return std::clamp(stepped, 0.25f, 1.0f);
}
[[nodiscard]] size_t ConfiguredGenerations() {
const u32 multiplier = std::clamp<u32>(Settings::values.frame_gen_multiplier.GetValue(),
LSFG_MIN_MULTIPLIER, LSFG_MAX_MULTIPLIER);
return multiplier - 1;
return Settings::FrameGenGenerations();
}
bool IsBlueFirst(VkFormat format) {
@@ -182,10 +197,15 @@ FrameGen::FrameGen(MemoryAllocator& memory_allocator_, Scheduler& scheduler_)
FrameGen::~FrameGen() = default;
void FrameGen::Process(const Device& device, Frame* frame, VkFormat format, bool generate) {
void FrameGen::Process(const Device& device, Frame* frame, VkFormat format, VkExtent2D guest_extent,
bool generate) {
generated = false;
if (unavailable || !Settings::values.frame_gen.GetValue()) {
if (unavailable || ConfiguredGenerations() == 0) {
if (chain) {
scheduler.Finish();
chain.reset();
}
return;
}
@@ -202,11 +222,15 @@ void FrameGen::Process(const Device& device, Frame* frame, VkFormat format, bool
}
}
peak_guest_extent.width = std::max(peak_guest_extent.width, guest_extent.width);
peak_guest_extent.height = std::max(peak_guest_extent.height, guest_extent.height);
const VkExtent2D extent{.width = frame->width, .height = frame->height};
const f32 flow_scale = ConfiguredFlowScale(peak_guest_extent, extent);
if (!chain || built_extent.width != extent.width || built_extent.height != extent.height ||
built_format != format || built_flow_scale != ConfiguredFlowScale() ||
built_format != format || built_flow_scale != flow_scale ||
built_generations != ConfiguredGenerations()) {
Rebuild(device, extent, format);
Rebuild(device, extent, format, flow_scale);
}
const u64 count = frame_count++;
@@ -232,7 +256,7 @@ void FrameGen::Process(const Device& device, Frame* frame, VkFormat format, bool
}
size_t FrameGen::WantedGenerations() const {
if (unavailable || !Settings::values.frame_gen.GetValue()) {
if (unavailable) {
return 0;
}
return ConfiguredGenerations();
@@ -254,11 +278,11 @@ void FrameGen::GenerateInto(const Device& device, Frame* destination, size_t gen
});
}
void FrameGen::Rebuild(const Device& device, VkExtent2D extent, VkFormat format) {
void FrameGen::Rebuild(const Device& device, VkExtent2D extent, VkFormat format, f32 flow_scale) {
scheduler.Finish();
chain.reset();
built_flow_scale = ConfiguredFlowScale();
built_flow_scale = flow_scale;
built_generations = ConfiguredGenerations();
chain.emplace(device, memory_allocator, *shaders, extent, format, built_flow_scale,
@@ -21,7 +21,8 @@ public:
explicit FrameGen(MemoryAllocator& memory_allocator, Scheduler& scheduler);
~FrameGen();
void Process(const Device& device, Frame* frame, VkFormat format, bool generate);
void Process(const Device& device, Frame* frame, VkFormat format, VkExtent2D guest_extent,
bool generate);
[[nodiscard]] size_t WantedGenerations() const;
@@ -30,7 +31,7 @@ public:
void GenerateInto(const Device& device, Frame* destination, size_t generation);
private:
void Rebuild(const Device& device, VkExtent2D extent, VkFormat format);
void Rebuild(const Device& device, VkExtent2D extent, VkFormat format, f32 flow_scale);
void DumpDebugImages(u64 count);
MemoryAllocator& memory_allocator;
@@ -38,6 +39,7 @@ private:
std::optional<LsfgShaders> shaders;
std::optional<LsfgChain> chain;
VkExtent2D peak_guest_extent{};
VkExtent2D built_extent{};
VkFormat built_format{VK_FORMAT_UNDEFINED};
f32 built_flow_scale{};
@@ -27,9 +27,6 @@ constexpr VkFormat LSFG_FLOW_FORMAT = VK_FORMAT_R8_UNORM;
constexpr VkFormat LSFG_MOTION_FORMAT = VK_FORMAT_R16G16B16A16_SFLOAT;
constexpr size_t LSFG_HISTORY_SLOTS = 3;
constexpr size_t LSFG_MIN_MULTIPLIER = 2;
constexpr size_t LSFG_MAX_MULTIPLIER = 4;
constexpr size_t LSFG_MAX_GENERATIONS = LSFG_MAX_MULTIPLIER - 1;
constexpr size_t LSFG_MAX_TARGETS = 7;
[[nodiscard]] constexpr f32 LsfgTimestamp(size_t generation, size_t generation_count) {
@@ -49,6 +49,21 @@ constexpr VkExtent2D CaptureImageSize{
.height = VideoCore::Capture::LinearHeight,
};
[[nodiscard]] VkExtent2D GuestExtent(std::span<const Tegra::FramebufferConfig> framebuffers) {
if (framebuffers.empty()) {
return VkExtent2D{};
}
const auto& framebuffer = framebuffers.front();
if (framebuffer.crop_rect.IsEmpty()) {
return VkExtent2D{.width = framebuffer.width, .height = framebuffer.height};
}
return VkExtent2D{
.width = static_cast<u32>(framebuffer.crop_rect.GetWidth()),
.height = static_cast<u32>(framebuffer.crop_rect.GetHeight()),
};
}
constexpr VkExtent3D CaptureImageExtent{
.width = VideoCore::Capture::LinearWidth,
.height = VideoCore::Capture::LinearHeight,
@@ -196,7 +211,8 @@ void RendererVulkan::Composite(std::span<const Tegra::FramebufferConfig> framebu
const size_t wanted = frame_gen.WantedGenerations();
const bool can_present_all = wanted > 0 && present_manager.AvailableExtraFrames() >= wanted;
frame_gen.Process(device, frame, swapchain.GetImageFormat(), can_present_all);
frame_gen.Process(device, frame, swapchain.GetImageFormat(), GuestExtent(framebuffers),
can_present_all);
const size_t generated_frames = frame_gen.GeneratedFrameCount();
for (size_t generation = 0; generation < generated_frames; ++generation) {
@@ -351,14 +351,11 @@ void PresentManager::SetImageCount() {
// We cannot have more than 7 images in flight at any given time.
// FRAMES_IN_FLIGHT is 8, and the cache TICKS_TO_DESTROY is 8.
// Mali drivers will give us 6.
const size_t generations =
Settings::values.frame_gen.GetValue()
? static_cast<size_t>(Settings::values.frame_gen_multiplier.GetValue()) - 1
: 0;
const size_t frames_per_composite = generations + 1;
image_count = std::min<size_t>(
std::max<size_t>(swapchain.GetImageCount() + generations, frames_per_composite * 2),
MAX_FRAMES_IN_FLIGHT);
const size_t generations = Settings::FrameGenGenerations();
const size_t queued_composites = Settings::values.frame_gen_queue_target.GetValue() + 1;
image_count =
std::clamp<size_t>((generations + 1) * queued_composites, swapchain.GetImageCount(),
MAX_FRAMES_IN_FLIGHT);
}
void PresentManager::CopyToSwapchain(Frame* frame) {