Compare commits

..

1 Commits

Author SHA1 Message Date
CamilleLaVey 677718d645 [TEST] Ported PSO optimizations from tiled-gpu-v2 2026-08-23 20:15:02 +00:00
69 changed files with 671 additions and 2269 deletions
@@ -27,7 +27,6 @@ enum class BooleanSetting(override val key: String) : AbstractBooleanSetting {
RENDERER_ASYNCHRONOUS_GPU_EMULATION("use_asynchronous_gpu_emulation"),
RENDERER_ASYNC_PRESENTATION("async_presentation"),
RENDERER_ASYNCHRONOUS_SHADERS("use_asynchronous_shaders"),
RENDERER_UNIFIED_MEMORY("use_unified_memory"),
RENDERER_REACTIVE_FLUSHING("use_reactive_flushing"),
ENABLE_BUFFER_HISTORY("enable_buffer_history"),
USE_OPTIMIZED_VERTEX_BUFFERS("use_optimized_vertex_buffers"),
@@ -624,7 +624,6 @@ abstract class SettingsItem(
IntSetting.FSR_SHARPENING_SLIDER,
titleId = R.string.fsr_sharpness,
descriptionId = R.string.fsr_sharpness_description,
max = 200,
units = "%"
)
)
@@ -789,13 +788,6 @@ abstract class SettingsItem(
descriptionId = R.string.renderer_asynchronous_shaders_description
)
)
put(
SwitchSetting(
BooleanSetting.RENDERER_UNIFIED_MEMORY,
titleId = R.string.renderer_unified_memory,
descriptionId = R.string.renderer_unified_memory_description
)
)
put(
SingleChoiceSetting(
IntSetting.FAST_GPU_TIME,
@@ -344,7 +344,6 @@ class SettingsFragmentPresenter(
add(BooleanSetting.FIX_BLOOM_EFFECTS.key)
add(BooleanSetting.EMULATE_BGR565.key)
add(BooleanSetting.RENDERER_ASYNCHRONOUS_SHADERS.key)
add(BooleanSetting.RENDERER_UNIFIED_MEMORY.key)
add(IntSetting.ANDROID_PIPELINE_WORKERS.key)
add(BooleanSetting.RENDERER_ASYNCHRONOUS_GPU_EMULATION.key)
add(BooleanSetting.RENDERER_ASYNC_PRESENTATION.key)
@@ -1182,7 +1182,7 @@ class EmulationFragment : Fragment(), SurfaceHolder.Callback {
container,
IntSetting.FSR_SHARPENING_SLIDER,
minValue = 0,
maxValue = 200,
maxValue = 100,
units = "%"
)
}
@@ -1107,6 +1107,7 @@
<string name="theme_mode_light">فاتح</string>
<string name="theme_mode_dark">داكن</string>
<string name="multiplier_none">لا شيء</string>
<!-- Black backgrounds theme -->
<string name="use_black_backgrounds">خلفيات سوداء</string>
@@ -991,6 +991,7 @@ Wirklich fortfahren?</string>
<string name="theme_mode_light">Hell</string>
<string name="theme_mode_dark">Dunkel</string>
<string name="multiplier_none">Keine</string>
<!-- Black backgrounds theme -->
<string name="use_black_backgrounds">Schwarze Hintergründe</string>
@@ -1092,6 +1092,7 @@
<string name="theme_mode_light">Claro</string>
<string name="theme_mode_dark">Oscuro</string>
<string name="multiplier_none">Nada</string>
<!-- Black backgrounds theme -->
<string name="use_black_backgrounds">Fondos oscuros</string>
@@ -808,6 +808,9 @@
<string name="multiplier_x4">x4</string>
<string name="multiplier_x8">x8</string>
<string name="multiplier_x16">x16</string>
<string name="multiplier_x32">x32</string>
<string name="multiplier_x64">x64</string>
<string name="multiplier_none">None</string>
<!-- Black backgrounds theme -->
<string name="use_black_backgrounds">پس‌زمینه مشکی</string>
@@ -1004,6 +1004,7 @@
<string name="theme_mode_light">Lumineux</string>
<string name="theme_mode_dark">Sombre</string>
<string name="multiplier_none">Aucun</string>
<!-- Black backgrounds theme -->
<string name="use_black_backgrounds">Arrière-plan noir</string>
@@ -935,6 +935,7 @@
<string name="theme_mode_light">Jasny</string>
<string name="theme_mode_dark">Ciemny</string>
<string name="multiplier_none">Brak</string>
<!-- Black backgrounds theme -->
<string name="use_black_backgrounds">Czarne tła</string>
@@ -891,6 +891,7 @@
<string name="theme_mode_light">Claro</string>
<string name="theme_mode_dark">Escuro</string>
<string name="multiplier_none">Nenhum</string>
<!-- Black backgrounds theme -->
<string name="use_black_backgrounds">Planos de fundo pretos</string>
@@ -1071,6 +1071,7 @@
<string name="theme_mode_light">Светлая</string>
<string name="theme_mode_dark">Темная</string>
<string name="multiplier_none">Отключено</string>
<!-- Black backgrounds theme -->
<string name="use_black_backgrounds">Чёрный фон</string>
@@ -1053,6 +1053,7 @@
<string name="theme_mode_light">Світла</string>
<string name="theme_mode_dark">Темна</string>
<string name="multiplier_none">Жодного</string>
<!-- Black backgrounds theme -->
<string name="use_black_backgrounds">Чорний фон</string>
@@ -1081,6 +1081,7 @@
<string name="theme_mode_light">浅色</string>
<string name="theme_mode_dark">深色</string>
<string name="multiplier_none"></string>
<!-- Black backgrounds theme -->
<string name="use_black_backgrounds">使用黑色背景</string>
@@ -1006,6 +1006,7 @@
<string name="theme_mode_light">淺色</string>
<string name="theme_mode_dark">深色</string>
<string name="multiplier_none"></string>
<!-- Black backgrounds theme -->
<string name="use_black_backgrounds">黑色背景</string>
+23 -12
View File
@@ -111,38 +111,43 @@
<item>1</item>
</integer-array>
<!-- VRAM USAGE MODE CHOICES -->
<string-array name="vramUsageMethodNames">
<item>@string/vram_usage_conservative</item>
<item>@string/vram_usage_aggressive</item>
</string-array>
<!-- VRAM USAGE MODE VALUES -->
<integer-array name="vramUsageMethodValues">
<item>0</item>
<item>1</item>
<item>0</item> <!-- Conservative -->
<item>1</item> <!-- Aggressive -->
</integer-array>
<!-- ASTC Decoding Method Choices -->
<string-array name="astcDecodingMethodNames">
<item>@string/accelerate_astc_cpu</item>
<item>@string/accelerate_astc_gpu</item>
<item>@string/accelerate_astc_async</item>
</string-array>
<!-- ASTC Decoding Method Values -->
<integer-array name="astcDecodingMethodValues">
<item>0</item>
<item>1</item>
<item>2</item>
<item>0</item> <!-- CPU -->
<item>1</item> <!-- GPU -->
<item>2</item> <!-- CPU Asynchronously -->
</integer-array>
<!-- NVDEC Emulation Choices -->
<string-array name="rendererNvdecNames">
<item>@string/nvdec_emulation_none</item>
<item>@string/nvdec_emulation_cpu</item>
<item>@string/nvdec_emulation_gpu</item>
<item>@string/nvdec_emulation_none</item> <!-- Off -->
<item>@string/nvdec_emulation_cpu</item> <!-- Cpu -->
<item>@string/nvdec_emulation_gpu</item> <!-- Gpu -->
</string-array>
<!-- NVDEC Emulation Values -->
<integer-array name="rendererNvdecValues">
<item>0</item>
<item>1</item>
<item>2</item>
<item>3</item> <!-- Off value -->
<item>1</item> <!-- CPU value -->
<item>2</item> <!-- GPU value -->
</integer-array>
<string-array name="rendererResolutionNames">
@@ -508,6 +513,9 @@
<item>@string/multiplier_x4</item>
<item>@string/multiplier_x8</item>
<item>@string/multiplier_x16</item>
<item>@string/multiplier_x32</item>
<item>@string/multiplier_x64</item>
<item>@string/multiplier_none</item>
</string-array>
<integer-array name="anisoValues">
<item>0</item>
@@ -516,6 +524,9 @@
<item>3</item>
<item>4</item>
<item>5</item>
<item>6</item>
<item>7</item>
<item>8</item>
</integer-array>
<string-array name="verticalAlignmentEntries">
@@ -585,8 +585,6 @@
<string name="rescale_hack_description">Enables a legacy handling for the rescale configuration pass for games by using a quick rescale path</string>
<string name="renderer_asynchronous_shaders">Use asynchronous shaders</string>
<string name="renderer_asynchronous_shaders_description">Compiles shaders asynchronously. This may reduce stutters but may also introduce glitches.</string>
<string name="renderer_unified_memory">Unified memory access</string>
<string name="renderer_unified_memory_description">Allows GPU write buffer readbacks directly into guest memory, skipping the CPU staging copy.</string>
<string name="gpu_unswizzle_settings">GPU Unswizzle Settings</string>
<string name="gpu_unswizzle_settings_description">Configure GPU-based texture unswizzling parameters or disable it entirely. Adjust these settings to balance performance and texture loading quality.</string>
<string name="gpu_unswizzle_enable">Enable GPU Unswizzle</string>
@@ -1248,6 +1246,9 @@
<string name="multiplier_x4" translatable="false">x4</string>
<string name="multiplier_x8" translatable="false">x8</string>
<string name="multiplier_x16" translatable="false">x16</string>
<string name="multiplier_x32" translatable="false">x32</string>
<string name="multiplier_x64" translatable="false">x64</string>
<string name="multiplier_none">None</string>
<!-- Black backgrounds theme -->
<string name="use_black_backgrounds">Black backgrounds</string>
+28 -22
View File
@@ -4,17 +4,34 @@
// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <fstream>
#include "common/heap_tracker.h"
#include "common/logging.h"
#include "common/memory_detect.h"
#include "common/assert.h"
namespace Common {
namespace {
s64 GetMaxPermissibleResidentMapCount() {
// Default value.
s64 value = 65530;
// Try to read how many mappings we can make.
std::ifstream s("/proc/sys/vm/max_map_count");
s >> value;
// Print, for debug.
LOG_INFO(HW_Memory, "Current maximum map count: {}", value);
// Allow 20000 maps for other code and to account for split inaccuracy.
return std::max<s64>(value - 20000, 0);
}
} // namespace
HeapTracker::HeapTracker(Common::HostMemory& buffer)
: m_buffer(buffer),
m_has_hardware_buffer_backing(!buffer.BackingHardwareBuffers().empty()),
m_max_resident_map_count(static_cast<s64>(GetPermissibleMapCount())) {}
: m_buffer(buffer), m_max_resident_map_count(GetMaxPermissibleResidentMapCount()) {}
HeapTracker::~HeapTracker() = default;
void HeapTracker::Map(size_t virtual_offset, size_t host_offset, size_t length,
@@ -68,8 +85,7 @@ void HeapTracker::Unmap(size_t virtual_offset, size_t size, bool is_separate_hea
// If resident, erase from resident map.
if (item->is_resident) {
m_resident_map_count -= this->HostMapCount(item->paddr, item->size);
ASSERT(m_resident_map_count >= 0);
ASSERT(--m_resident_map_count >= 0);
m_resident_mappings.erase(m_resident_mappings.iterator_to(*item));
}
@@ -175,7 +191,7 @@ bool HeapTracker::DeferredMapSeparateHeap(size_t virtual_offset) {
// This map is now resident.
it->is_resident = true;
m_resident_map_count += this->HostMapCount(it->paddr, it->size);
m_resident_map_count++;
m_resident_mappings.insert(*it);
}
@@ -197,17 +213,17 @@ void HeapTracker::RebuildSeparateHeapAddressSpace() {
// Despite being worse in theory, this has proven to be better in practice than more
// regularly dumping a smaller amount, because it significantly reduces average case
// lock contention.
s64 const desired_count = (std::min)(m_resident_map_count, m_max_resident_map_count) / 2;
std::size_t const desired_count = (std::min)(m_resident_map_count, m_max_resident_map_count) / 2;
std::size_t const evict_count = m_resident_map_count - desired_count;
auto it = m_resident_mappings.begin();
while (m_resident_map_count > desired_count && it != m_resident_mappings.end()) {
for (size_t i = 0; i < evict_count && it != m_resident_mappings.end(); i++) {
// Unmark and unmap.
it->is_resident = false;
m_buffer.Unmap(it->vaddr, it->size, false);
// Advance.
m_resident_map_count -= this->HostMapCount(it->paddr, it->size);
ASSERT(m_resident_map_count >= 0);
ASSERT(--m_resident_map_count >= 0);
it = m_resident_mappings.erase(it);
}
}
@@ -229,7 +245,6 @@ void HeapTracker::SplitHeapMapLocked(VAddr offset) {
// Cache the original values.
auto* const left = std::addressof(*it);
const size_t orig_size = left->size;
const s64 orig_host_map_count = this->HostMapCount(left->paddr, orig_size);
// Adjust the left map.
const size_t left_size = offset - left->vaddr;
@@ -251,20 +266,11 @@ void HeapTracker::SplitHeapMapLocked(VAddr offset) {
// If resident, also insert into resident map.
if (right->is_resident) {
m_resident_map_count += this->HostMapCount(left->paddr, left->size) +
this->HostMapCount(right->paddr, right->size) -
orig_host_map_count;
m_resident_map_count++;
m_resident_mappings.insert(*right);
}
}
s64 HeapTracker::HostMapCount(PAddr paddr, size_t size) const {
if (!m_has_hardware_buffer_backing) {
return size != 0 ? 1 : 0;
}
return static_cast<s64>(m_buffer.BackingMapCount(paddr, size));
}
HeapTracker::AddrTree::iterator HeapTracker::GetNearestHeapMapLocked(VAddr offset) {
const SeparateHeapMap key{
.vaddr = offset,
-6
View File
@@ -1,6 +1,3 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
@@ -85,13 +82,10 @@ private:
AddrTree::iterator GetNearestHeapMapLocked(VAddr offset);
s64 HostMapCount(PAddr paddr, size_t size) const;
void RebuildSeparateHeapAddressSpace();
private:
Common::HostMemory& m_buffer;
const bool m_has_hardware_buffer_backing;
const s64 m_max_resident_map_count;
std::shared_mutex m_rebuild_lock{};
+6 -389
View File
@@ -51,65 +51,14 @@
#endif // ^^^ POSIX ^^^
#include <atomic>
#include <mutex>
#include <random>
#include <vector>
#include "common/alignment.h"
#include "common/assert.h"
#include "common/free_region_manager.h"
#include "common/host_memory.h"
#include "common/logging.h"
#include "common/memory_detect.h"
#include "common/settings.h"
#ifdef __ANDROID__
#include <cerrno>
#include <dlfcn.h>
#include <sys/ioctl.h>
#include <android/hardware_buffer.h>
namespace {
struct NativeHandle {
int version;
int numFds;
int numInts;
int data[1];
};
using PFN_AHardwareBuffer_getNativeHandle = const NativeHandle* (*)(const AHardwareBuffer*);
PFN_AHardwareBuffer_getNativeHandle ResolveGetNativeHandle() {
void* const lib = dlopen("libnativewindow.so", RTLD_NOW);
if (lib == nullptr) {
return nullptr;
}
return reinterpret_cast<PFN_AHardwareBuffer_getNativeHandle>(
dlsym(lib, "AHardwareBuffer_getNativeHandle"));
}
struct DmaBufSync {
u64 flags;
};
constexpr u64 DmaBufSyncRead = 1ULL << 0;
constexpr u64 DmaBufSyncWrite = 1ULL << 1;
constexpr u64 DmaBufSyncStart = 0ULL << 2;
constexpr u64 DmaBufSyncEnd = 1ULL << 2;
void SyncDmaBufCpuAccess(int fd, u64 phase) {
DmaBufSync sync{.flags = phase | DmaBufSyncRead | DmaBufSyncWrite};
while (ioctl(fd, _IOW('b', 0, DmaBufSync), &sync) != 0) {
if (errno != EINTR) {
return;
}
}
}
} // namespace
#endif
#if defined(__ANDROID__) && __ANDROID_API__ < 30
#include <sys/syscall.h>
@@ -126,12 +75,6 @@ namespace Common {
[[maybe_unused]] constexpr size_t PageAlignment = 0x1000;
[[maybe_unused]] constexpr size_t HugePageSize = 0x200000;
static std::atomic<u64> committed_backing_size{};
u64 GetCommittedBackingSize() noexcept {
return committed_backing_size.load(std::memory_order_relaxed);
}
#ifdef _WIN32
// Manually imported for MinGW compatibility
@@ -180,7 +123,7 @@ static void GetFuncAddress(Common::DynamicLibrary& dll, const char* name, T& pfn
class HostMemory::Impl {
public:
explicit Impl(size_t backing_size_, size_t virtual_size_, size_t)
explicit Impl(size_t backing_size_, size_t virtual_size_)
: backing_size{backing_size_}
, virtual_size{virtual_size_}
, process{GetCurrentProcess()}
@@ -286,10 +229,6 @@ public:
UNREACHABLE();
}
bool IsBackingShared() const noexcept {
return true;
}
const size_t backing_size; ///< Size of the backing memory in bytes
const size_t virtual_size; ///< Size of the virtual address placeholder in bytes
@@ -562,10 +501,9 @@ static int shm_open_anon(int flags, mode_t mode) {
class HostMemory::Impl {
public:
explicit Impl(size_t backing_size_, size_t virtual_size_, size_t preferred_offset_)
explicit Impl(size_t backing_size_, size_t virtual_size_)
: backing_size{backing_size_}
, virtual_size{virtual_size_}
, preferred_offset{preferred_offset_}
{}
bool Init() {
@@ -605,15 +543,10 @@ public:
LOG_WARNING(Common_Memory, "Using private mappings instead of shared ones");
backing_base = static_cast<u8*>(mmap(nullptr, backing_size, PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_PRIVATE, -1, 0));
if (fd > 0) {
fd = -1;
close(fd);
}
fd = -1;
} else {
#ifdef __ANDROID__
if (InitAhbBacking()) {
return InitVirtual();
}
#endif
backing_base = static_cast<u8*>(mmap(nullptr, backing_size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0));
}
if (backing_base == MAP_FAILED) {
@@ -621,10 +554,7 @@ public:
return false;
}
return InitVirtual();
}
bool InitVirtual() {
// Virtual memory initialization
virtual_base = virtual_map_base = static_cast<u8*>(ChooseVirtualBase(virtual_size));
if (virtual_base == MAP_FAILED) {
LOG_CRITICAL(HW_Memory, "mmap failed: {}", strerror(errno));
@@ -637,240 +567,6 @@ public:
return true;
}
#ifdef __ANDROID__
static AHardwareBuffer_Desc MakeBlobDesc(size_t len) {
return AHardwareBuffer_Desc{
.width = static_cast<u32>(len),
.height = 1,
.layers = 1,
.format = AHARDWAREBUFFER_FORMAT_BLOB,
.usage = AHARDWAREBUFFER_USAGE_CPU_READ_OFTEN |
AHARDWAREBUFFER_USAGE_CPU_WRITE_OFTEN |
AHARDWAREBUFFER_USAGE_GPU_DATA_BUFFER,
.stride = 0,
.rfu0 = 0,
.rfu1 = 0,
};
}
static bool ProbeAhbBacking(PFN_AHardwareBuffer_getNativeHandle get_native_handle) {
const AHardwareBuffer_Desc desc = MakeBlobDesc(PageAlignment * 2);
AHardwareBuffer* buffer{};
if (AHardwareBuffer_allocate(&desc, &buffer) != 0 || buffer == nullptr) {
return false;
}
const NativeHandle* const handle = get_native_handle(buffer);
if (handle == nullptr || handle->numFds < 1) {
AHardwareBuffer_release(buffer);
return false;
}
const int probe_fd = handle->data[0];
bool ok = true;
const auto try_map = [&](int prot, off_t offset) {
if (!ok) {
return;
}
void* const ptr = mmap(nullptr, PageAlignment, prot, MAP_SHARED, probe_fd, offset);
if (ptr == MAP_FAILED) {
ok = false;
return;
}
munmap(ptr, PageAlignment);
};
try_map(PROT_READ | PROT_WRITE, 0);
try_map(PROT_READ | PROT_WRITE, static_cast<off_t>(PageAlignment));
#ifdef ARCHITECTURE_arm64
try_map(PROT_READ | PROT_EXEC, 0);
#endif
AHardwareBuffer_release(buffer);
return ok;
}
size_t ComputeAhbBudget(size_t window_size) const {
const u64 total_physical = Common::GetMemInfo().TotalPhysicalMemory;
constexpr u64 BaselineFootprint = 6ULL << 30;
if (total_physical <= BaselineFootprint) {
return 0;
}
const u64 permissible_maps = Common::GetPermissibleMapCount();
if (permissible_maps == 0) {
return 0;
}
u64 budget = (total_physical - BaselineFootprint) / 2;
constexpr u64 MapSlotsPerWindow = 64;
const u64 affordable_windows = permissible_maps / MapSlotsPerWindow;
budget = (std::min)(budget, affordable_windows * window_size);
budget = (std::min)(budget, static_cast<u64>(backing_size));
budget = Common::AlignDown(budget, window_size);
constexpr u64 MinimumBudget = 256ULL << 20;
if (budget < MinimumBudget) {
return 0;
}
return static_cast<size_t>(budget);
}
bool InitAhbBacking() {
if (!Settings::values.use_unified_memory.GetValue()) {
return false;
}
static const PFN_AHardwareBuffer_getNativeHandle get_native_handle =
ResolveGetNativeHandle();
if (get_native_handle == nullptr) {
return false;
}
constexpr size_t window_size = 256ULL << 20;
const size_t budget = ComputeAhbBudget(window_size);
if (budget == 0) {
return false;
}
if (!ProbeAhbBacking(get_native_handle)) {
return false;
}
const size_t aligned_backing = Common::AlignDown(backing_size, window_size);
const size_t max_windows = (std::min)(budget, aligned_backing) / window_size;
std::vector<AHardwareBuffer*> buffers;
std::vector<int> buffer_fds;
const auto cleanup = [&] {
for (AHardwareBuffer* buffer : buffers) {
AHardwareBuffer_release(buffer);
}
buffers.clear();
buffer_fds.clear();
};
for (size_t i = 0; i < max_windows; ++i) {
const AHardwareBuffer_Desc desc = MakeBlobDesc(window_size);
AHardwareBuffer* buffer{};
if (AHardwareBuffer_allocate(&desc, &buffer) != 0 || buffer == nullptr) {
break;
}
const NativeHandle* const handle = get_native_handle(buffer);
if (handle == nullptr || handle->numFds < 1) {
AHardwareBuffer_release(buffer);
break;
}
const int buffer_fd = handle->data[0];
const off_t buffer_len = lseek(buffer_fd, 0, SEEK_END);
if (buffer_len < static_cast<off_t>(window_size)) {
AHardwareBuffer_release(buffer);
break;
}
buffers.push_back(buffer);
buffer_fds.push_back(buffer_fd);
}
const size_t num_windows = buffers.size();
if (num_windows == 0) {
return false;
}
const size_t region_size = num_windows * window_size;
const size_t region_base = Common::AlignDown(
(std::min)(preferred_offset, aligned_backing - region_size), window_size);
u8* const base = static_cast<u8*>(mmap(nullptr, backing_size, PROT_NONE,
MAP_PRIVATE | MAP_ANONYMOUS | MAP_NORESERVE, -1, 0));
if (base == MAP_FAILED) {
cleanup();
return false;
}
const auto map_over_reservation = [&](size_t offset, size_t len, int map_fd,
off_t map_offset) {
if (len == 0) {
return true;
}
if (mmap(base + offset, len, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_FIXED, map_fd,
map_offset) == MAP_FAILED) {
munmap(base, backing_size);
cleanup();
return false;
}
return true;
};
if (!map_over_reservation(0, region_base, fd, 0)) {
return false;
}
for (size_t i = 0; i < num_windows; ++i) {
if (!map_over_reservation(region_base + i * window_size, window_size, buffer_fds[i],
0)) {
return false;
}
}
const size_t tail_offset = region_base + region_size;
if (!map_over_reservation(tail_offset, backing_size - tail_offset, fd,
static_cast<off_t>(tail_offset))) {
return false;
}
backing_base = base;
ahb_windows = std::move(buffers);
ahb_fds = std::move(buffer_fds);
ahb_window_size = window_size;
ahb_base = region_base;
ahb_bytes = region_size;
committed_backing_size.store(region_size, std::memory_order_relaxed);
for (const int window_fd : ahb_fds) {
SyncDmaBufCpuAccess(window_fd, DmaBufSyncStart);
}
return true;
}
void MapBackingRange(size_t virtual_offset, size_t host_offset, size_t length, int prot_flags) {
while (length > 0) {
int map_fd = fd;
off_t map_offset = static_cast<off_t>(host_offset);
size_t chunk = length;
if (host_offset < ahb_base) {
chunk = (std::min)(chunk, ahb_base - host_offset);
} else if (host_offset < ahb_base + ahb_bytes) {
const size_t relative = host_offset - ahb_base;
const size_t window = relative / ahb_window_size;
const size_t local = relative % ahb_window_size;
map_fd = ahb_fds[window];
map_offset = static_cast<off_t>(local);
chunk = (std::min)(chunk, ahb_window_size - local);
}
void* const ret = mmap(virtual_base + virtual_offset, chunk, prot_flags,
MAP_SHARED | MAP_FIXED, map_fd, map_offset);
ASSERT_MSG(ret != MAP_FAILED, "mmap: {}", strerror(errno));
virtual_offset += chunk;
host_offset += chunk;
length -= chunk;
}
}
size_t BackingMapCount(size_t host_offset, size_t length) const noexcept {
if (length == 0) {
return 0;
}
if (ahb_bytes == 0) {
return 1;
}
size_t count = 0;
while (length > 0) {
size_t chunk = length;
if (host_offset < ahb_base) {
chunk = (std::min)(chunk, ahb_base - host_offset);
} else if (host_offset < ahb_base + ahb_bytes) {
const size_t local = (host_offset - ahb_base) % ahb_window_size;
chunk = (std::min)(chunk, ahb_window_size - local);
}
host_offset += chunk;
length -= chunk;
++count;
}
return count;
}
std::span<AHardwareBuffer* const> AhbWindows() const noexcept {
return ahb_windows;
}
size_t AhbWindowSize() const noexcept {
return ahb_bytes != 0 ? ahb_window_size : 0;
}
size_t AhbBase() const noexcept {
return ahb_base;
}
#endif
~Impl() {
Release();
}
@@ -891,12 +587,6 @@ public:
#ifdef ARCHITECTURE_arm64
if (True(perms & MemoryPermission::Execute))
prot_flags |= PROT_EXEC;
#endif
#ifdef __ANDROID__
if (ahb_bytes != 0) {
MapBackingRange(virtual_offset, host_offset, length, prot_flags);
return;
}
#endif
int flags = (fd >= 0 ? MAP_SHARED : MAP_PRIVATE) | MAP_FIXED;
void* ret = mmap(virtual_base + virtual_offset, length, prot_flags, flags, fd, host_offset);
@@ -942,18 +632,8 @@ public:
virtual_base = nullptr;
}
bool IsBackingShared() const noexcept {
#ifdef __ANDROID__
if (ahb_bytes != 0) {
return true;
}
#endif
return fd >= 0;
}
const size_t backing_size; ///< Size of the backing memory in bytes
const size_t virtual_size; ///< Size of the virtual address placeholder in bytes
const size_t preferred_offset;
u8* backing_base{reinterpret_cast<u8*>(MAP_FAILED)};
u8* virtual_base{reinterpret_cast<u8*>(MAP_FAILED)};
@@ -976,21 +656,6 @@ private:
int ret = close(fd);
ASSERT_MSG(ret == 0, "close failed: {}", strerror(errno));
}
#ifdef __ANDROID__
for (const int window_fd : ahb_fds) {
SyncDmaBufCpuAccess(window_fd, DmaBufSyncEnd);
}
for (AHardwareBuffer* buffer : ahb_windows) {
AHardwareBuffer_release(buffer);
}
ahb_windows.clear();
ahb_fds.clear();
if (ahb_bytes != 0) {
committed_backing_size.store(0, std::memory_order_relaxed);
ahb_bytes = 0;
}
#endif
}
void AdjustMap(size_t* virtual_offset, size_t* length) {
@@ -1016,19 +681,11 @@ private:
int fd{-1}; // memfd file descriptor, -1 is the error value of memfd_create
FreeRegionManager free_manager{};
#ifdef __ANDROID__
std::vector<AHardwareBuffer*> ahb_windows;
std::vector<int> ahb_fds;
size_t ahb_window_size{};
size_t ahb_base{};
size_t ahb_bytes{};
#endif
};
#endif // ^^^ POSIX ^^^
HostMemory::HostMemory(size_t backing_size_, size_t virtual_size_, size_t preferred_offset_)
HostMemory::HostMemory(size_t backing_size_, size_t virtual_size_)
: backing_size(backing_size_)
, virtual_size(virtual_size_)
{
@@ -1040,7 +697,7 @@ HostMemory::HostMemory(size_t backing_size_, size_t virtual_size_, size_t prefer
#else
// Try to allocate a fastmem arena.
// The implementation will fail with std::bad_alloc on errors.
impl = std::make_unique<HostMemory::Impl>(AlignUp(backing_size, PageAlignment), AlignUp(virtual_size, PageAlignment) + HugePageSize, preferred_offset_);
impl = std::make_unique<HostMemory::Impl>(AlignUp(backing_size, PageAlignment), AlignUp(virtual_size, PageAlignment) + HugePageSize);
if (impl->Init()) {
backing_base = impl->backing_base;
virtual_base = impl->virtual_base;
@@ -1110,46 +767,6 @@ void HostMemory::ClearBackingRegion(size_t physical_offset, size_t length, u32 f
std::memset(backing_base + physical_offset, fill_value, length);
}
std::span<AHardwareBuffer* const> HostMemory::BackingHardwareBuffers() const noexcept {
#ifdef __ANDROID__
return impl ? impl->AhbWindows() : std::span<AHardwareBuffer* const>{};
#else
return {};
#endif
}
size_t HostMemory::BackingMapCount(size_t host_offset, size_t length) const noexcept {
#ifdef __ANDROID__
return impl ? impl->BackingMapCount(host_offset, length) : (length != 0 ? 1 : 0);
#else
return length != 0 ? 1 : 0;
#endif
}
size_t HostMemory::BackingHardwareBufferWindowSize() const noexcept {
#ifdef __ANDROID__
return impl ? impl->AhbWindowSize() : 0;
#else
return 0;
#endif
}
bool HostMemory::IsBackingShared() const noexcept {
#if defined(__OPENORBIS__) || defined(__managarm__)
return false;
#else
return impl && impl->IsBackingShared();
#endif
}
size_t HostMemory::BackingHardwareBufferBase() const noexcept {
#ifdef __ANDROID__
return impl ? impl->AhbBase() : 0;
#else
return 0;
#endif
}
void HostMemory::EnableDirectMappedAddress() {
#if !(defined(__OPENORBIS__) || defined(__managarm__))
if (impl) {
+1 -20
View File
@@ -8,17 +8,12 @@
#include <memory>
#include <optional>
#include <span>
#include "common/common_funcs.h"
#include "common/common_types.h"
#include "common/virtual_buffer.h"
struct AHardwareBuffer;
namespace Common {
[[nodiscard]] u64 GetCommittedBackingSize() noexcept;
enum class MemoryPermission : u32 {
Read = 1 << 0,
Write = 1 << 1,
@@ -33,7 +28,7 @@ DECLARE_ENUM_FLAG_OPERATORS(MemoryPermission)
*/
class HostMemory {
public:
explicit HostMemory(size_t backing_size_, size_t virtual_size_, size_t preferred_offset_ = 0);
explicit HostMemory(size_t backing_size_, size_t virtual_size_);
~HostMemory();
/**
@@ -67,20 +62,6 @@ public:
return backing_base;
}
[[nodiscard]] size_t BackingSize() const noexcept {
return backing_size;
}
[[nodiscard]] size_t BackingMapCount(size_t host_offset, size_t length) const noexcept;
[[nodiscard]] std::span<AHardwareBuffer* const> BackingHardwareBuffers() const noexcept;
[[nodiscard]] size_t BackingHardwareBufferWindowSize() const noexcept;
[[nodiscard]] size_t BackingHardwareBufferBase() const noexcept;
[[nodiscard]] bool IsBackingShared() const noexcept;
[[nodiscard]] u8* VirtualBasePointer() noexcept {
return virtual_base;
}
+45 -56
View File
@@ -39,19 +39,6 @@
namespace Common::Log {
/// @brief A log entry. Log entries are store in a structured format to permit more varied output
/// formatting on different frontends, as well as facilitating filtering and aggregation.
struct Entry {
char const* message = nullptr;
size_t message_len = 0;
std::chrono::microseconds timestamp;
Class log_class{};
Level log_level{};
const char* filename = nullptr;
const char* function = nullptr;
uint32_t line_num = 0;
};
namespace {
/// @brief Returns the name of the passed log class as a C-string. Subclasses are separated by periods
@@ -83,6 +70,8 @@ const char* GetLevelName(Level log_level) {
}
}
}
// Some IDEs prefer <file>:<line> instead, so let's just do that :)
std::string FormatLogMessage(const Entry& entry) noexcept {
if (!entry.filename) return "";
@@ -90,9 +79,10 @@ std::string FormatLogMessage(const Entry& entry) noexcept {
auto const time_fractional = uint32_t(entry.timestamp.count() % 1000000);
auto const class_name = GetLogClassName(entry.log_class);
auto const level_name = GetLevelName(entry.log_level);
return fmt::format("[{:4d}.{:06d}] {} <{}> {}:{}:{}: {}\n", time_seconds, time_fractional, class_name, level_name, entry.filename, entry.line_num, entry.function, entry.message);
return fmt::format("[{:4d}.{:06d}] {} <{}> {}:{}:{}: {}", time_seconds, time_fractional, class_name, level_name, entry.filename, entry.line_num, entry.function, entry.message);
}
namespace {
template <typename It>
Level GetLevelByName(const It begin, const It end) {
for (u32 i = 0; i < u32(Level::Count); ++i) {
@@ -137,6 +127,25 @@ bool ParseFilterRule(Filter& instance, Iterator begin, Iterator end) {
instance.SetClassLevel(log_class, level);
return true;
}
} // Anonymous namespace
void Filter::ParseFilterString(std::string_view filter_view) {
auto clause_begin = filter_view.cbegin();
while (clause_begin != filter_view.cend()) {
auto clause_end = std::find(clause_begin, filter_view.cend(), ' ');
// If clause isn't empty
if (clause_end != clause_begin) {
ParseFilterRule(*this, clause_begin, clause_end);
}
if (clause_end != filter_view.cend()) {
// Skip over the whitespace
++clause_end;
}
clause_begin = clause_end;
}
}
namespace {
/// @brief Trims up to and including the last of ../, ..\, src/, src\ in a string
/// do not be fooled this isn't generating new strings on .rodata :)
@@ -199,7 +208,7 @@ struct ColorConsoleBackend final : public Backend {
}());
SetConsoleTextAttribute(console_handle, color);
auto const df = GetDirectFormatArgs(entry);
std::fprintf(stdout, CCB_PRINTF_FMT "\n", df.time_seconds, df.time_fractional, df.class_name, df.level_name, entry.filename, entry.line_num, entry.function, entry.message);
std::fprintf(stdout, CCB_PRINTF_FMT "\n", df.time_seconds, df.time_fractional, df.class_name, df.level_name, entry.filename, entry.line_num, entry.function, entry.message.c_str());
}
}
void Flush() noexcept override {}
@@ -211,24 +220,22 @@ struct ColorConsoleBackend final : public Backend {
~ColorConsoleBackend() noexcept override {}
void Write(const Entry& entry) noexcept override {
if (enabled) {
#define ESC "\x1b"
auto const color_str = [&entry]() -> const char* {
switch (entry.log_level) {
case Level::Debug: return "[0;36m"; // Cyan
case Level::Info: return "[0;37m"; // Bright gray
case Level::Warning: return "[1;33m"; // Bright yellow
case Level::Error: return "[1;31m"; // Bright red
case Level::Critical: return "[1;35m"; // Bright magenta
default: return "[1;30m"; // Grey
#define CCB_MAKE_COLOR_FMT(X) ESC X CCB_PRINTF_FMT ESC "[0m\n"
case Level::Debug: return CCB_MAKE_COLOR_FMT("[0;36m"); // Cyan
case Level::Info: return CCB_MAKE_COLOR_FMT("[0;37m"); // Bright gray
case Level::Warning: return CCB_MAKE_COLOR_FMT("[1;33m"); // Bright yellow
case Level::Error: return CCB_MAKE_COLOR_FMT("[1;31m"); // Bright red
case Level::Critical: return CCB_MAKE_COLOR_FMT("[1;35m"); // Bright magenta
default: return CCB_MAKE_COLOR_FMT("[1;30m"); // Grey
#undef CCB_MAKE_COLOR_FMT
}
}();
auto const df = GetDirectFormatArgs(entry);
// more restrictive, because take for example this simple prelude:
// [ 50.872256] Config <Info> common/settings.cpp:142:LogSettings:
char buffer[128];
auto result = fmt::format_to_n(buffer, sizeof(buffer) - 1, "\x1b{}[{:4d}.{:06d}] {} <{}> {}:{}:{}: ", color_str, df.time_seconds, df.time_fractional, df.class_name, df.level_name, entry.filename, entry.line_num, entry.function, entry.message);
std::fwrite(buffer, 1, (std::min)(sizeof(buffer) - 1, result.size), stdout);
std::fwrite(entry.message, 1, entry.message_len, stdout);
std::fwrite("\x1b[0m\n", 1, sizeof("\x1b[0m\n"), stdout);
std::fprintf(stdout, color_str, df.time_seconds, df.time_fractional, df.class_name, df.level_name, entry.filename, entry.line_num, entry.function, entry.message.c_str());
#undef ESC
}
}
void Flush() noexcept override {}
@@ -239,7 +246,7 @@ struct ColorConsoleBackend final : public Backend {
#ifndef __OPENORBIS__
/// @brief Backend that writes to a file passed into the constructor
struct FileBackend final : public Backend {
explicit FileBackend(const std::filesystem::path filename) noexcept {
explicit FileBackend(const std::filesystem::path& filename) noexcept {
auto old_filename = filename;
old_filename += ".old.txt";
// Existence checks are done within the functions themselves.
@@ -254,7 +261,7 @@ struct FileBackend final : public Backend {
if (!enabled)
return;
auto message = FormatLogMessage(entry);
auto message = FormatLogMessage(entry).append(1, '\n');
#ifndef __ANDROID__
if (Settings::values.censor_username.GetValue()) {
// This must be a static otherwise it would get checked on EVERY
@@ -262,7 +269,8 @@ struct FileBackend final : public Backend {
static std::string username = []() -> std::string {
// in order of precedence
// LOGNAME usually works on UNIX, USERNAME on Windows
// Some UNIX systems suck and don't use LOGNAME so we also need USER :(
// Some UNIX systems suck and don't use LOGNAME so we also
// need USER :(
for (auto const var : { "LOGNAME", "USERNAME", "USER", })
if (auto const s = ::getenv(var); s != nullptr)
return std::string{s};
@@ -272,7 +280,7 @@ struct FileBackend final : public Backend {
boost::replace_all(message, username, "user");
}
#endif
bytes_written += file->WriteSpan(std::span<const char>{message.begin(), message.end()});
bytes_written += file->WriteString(message);
// Option to log each line rather than 4k buffers
if (Settings::values.log_flush_line.GetValue())
@@ -300,13 +308,14 @@ private:
bool enabled = true;
};
#endif
#ifdef _WIN32
/// @brief Backend that writes to Visual Studio's output window
struct DebuggerBackend final : public Backend {
explicit DebuggerBackend() noexcept = default;
~DebuggerBackend() noexcept override = default;
void Write(const Entry& entry) noexcept override {
::OutputDebugStringW(UTF8ToUTF16W(FormatLogMessage(entry)).c_str());
::OutputDebugStringW(UTF8ToUTF16W(FormatLogMessage(entry).append(1, '\n')).c_str());
}
void Flush() noexcept override {}
};
@@ -329,7 +338,7 @@ struct LogcatBackend : public Backend {
}
}();
auto const df = GetDirectFormatArgs(entry);
__android_log_print(android_log_priority, "YuzuNative", CCB_PRINTF_FMT, df.time_seconds, df.time_fractional, df.class_name, df.level_name, entry.filename, entry.line_num, entry.function, entry.message);
__android_log_print(android_log_priority, "YuzuNative", CCB_PRINTF_FMT, df.time_seconds, df.time_fractional, df.class_name, df.level_name, entry.filename, entry.line_num, entry.function, entry.message.c_str());
}
void Flush() noexcept override {}
};
@@ -368,23 +377,7 @@ struct Impl {
#endif
std::chrono::steady_clock::time_point time_origin{std::chrono::steady_clock::now()};
};
} // Anonymous namespace
void Filter::ParseFilterString(std::string_view filter_view) {
auto clause_begin = filter_view.cbegin();
while (clause_begin < filter_view.cend()) {
auto clause_end = std::find(clause_begin, filter_view.cend(), ' ');
// If clause isn't empty
if (clause_end != clause_begin) {
ParseFilterRule(*this, clause_begin, clause_end);
}
if (clause_end != filter_view.cend()) {
// Skip over the whitespace
++clause_end;
}
clause_begin = clause_end;
}
}
} // namespace
// Constructor shall NOT depend upon Settings() or whatever
// it's ran at global static ctor() time... so BE CAREFUL MFER!
@@ -425,14 +418,10 @@ void SetColorConsoleBackendEnabled(bool enabled) {
void FmtLogMessageImpl(Class log_class, Level log_level, const char* filename, unsigned int line_num, const char* function, fmt::string_view format, const fmt::format_args& args) {
if (logging_instance && logging_instance->filter.CheckMessage(log_class, log_level)) {
char buffer[BUFSIZ];
auto result = fmt::vformat_to_n(buffer, sizeof(buffer) - 1, format, args);
buffer[result.size] = '\0';
auto const flush = ::Settings::values.log_flush_line.GetValue();
logging_instance->ForEachBackend([=](Backend& backend) {
backend.Write(Entry{
.message = buffer,
.message_len = (std::min)(sizeof(buffer) - 1, result.size),
.message = fmt::vformat(format, args),
.timestamp = std::chrono::duration_cast<std::chrono::microseconds>(std::chrono::steady_clock::now() - logging_instance->time_origin),
.log_class = log_class,
.log_level = log_level,
+21
View File
@@ -140,4 +140,25 @@ void Stop();
void SetGlobalFilter(const Filter& filter);
void SetColorConsoleBackendEnabled(bool enabled);
/// @brief A log entry. Log entries are store in a structured format to permit more varied output
/// formatting on different frontends, as well as facilitating filtering and aggregation.
struct Entry {
std::string message;
std::chrono::microseconds timestamp;
Class log_class{};
Level log_level{};
const char* filename = nullptr;
const char* function = nullptr;
unsigned int line_num = 0;
};
/// Formats a log entry into the provided text buffer.
std::string FormatLogMessage(const Entry& entry) noexcept;
/// Prints the same message as `PrintMessage`, but colored according to the severity level.
void PrintColoredMessage(const Entry& entry) noexcept;
/// Formats and prints a log entry to the android logcat.
void PrintMessageToLogcat(const Entry& entry) noexcept;
} // namespace Common::Log
-28
View File
@@ -1,6 +1,3 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
@@ -20,9 +17,6 @@
#endif
#endif
#include <cstdio>
#include <cstdlib>
#include "common/memory_detect.h"
namespace Common {
@@ -75,26 +69,4 @@ const MemoryInfo& GetMemInfo() {
return mem_info;
}
u64 GetPermissibleMapCount() {
constexpr u64 DefaultMapCount = 65530;
constexpr u64 ReservedMaps = 20000;
u64 count = DefaultMapCount;
#ifdef __linux__
if (std::FILE* const file = std::fopen("/proc/sys/vm/max_map_count", "re")) {
char line[32];
if (std::fgets(line, sizeof(line), file) != nullptr) {
const u64 parsed = std::strtoull(line, nullptr, 10);
if (parsed != 0) {
count = parsed;
}
}
std::fclose(file);
}
#endif
if (count <= ReservedMaps) {
return 0;
}
return count - ReservedMaps;
}
} // namespace Common
-5
View File
@@ -1,6 +1,3 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
@@ -21,6 +18,4 @@ struct MemoryInfo {
*/
[[nodiscard]] const MemoryInfo& GetMemInfo();
[[nodiscard]] u64 GetPermissibleMapCount();
} // namespace Common
-3
View File
@@ -645,9 +645,6 @@ struct Values {
SwitchableSetting<bool> use_asynchronous_shaders{linkage, false, "use_asynchronous_shaders",
Category::RendererHacks};
SwitchableSetting<bool> use_unified_memory{linkage, false, "use_unified_memory",
Category::RendererHacks};
SwitchableSetting<GpuUnswizzleSize> gpu_unswizzle_texture_size{linkage,
GpuUnswizzleSize::Large,
"gpu_unswizzle_texture_size",
+1 -1
View File
@@ -128,7 +128,7 @@ ENUM(TimeZone, Auto, Default, Cet, Cst6Cdt, Cuba, Eet, Egypt, Eire, Est, Est5Edt
GmtPlusZero, GmtMinusZero, GmtZero, Greenwich, Hongkong, Hst, Iceland, Iran, Israel, Jamaica,
Japan, Kwajalein, Libya, Met, Mst, Mst7Mdt, Navajo, Nz, NzChat, Poland, Portugal, Prc, Pst8Pdt,
Roc, Rok, Singapore, Turkey, Uct, Universal, Utc, WSu, Wet, Zulu);
ENUM(AnisotropyMode, Automatic, Default, X2, X4, X8, X16);
ENUM(AnisotropyMode, Automatic, Default, X2, X4, X8, X16, X32, X64, None);
ENUM(AstcDecodeMode, Cpu, Gpu, CpuAsynchronous);
ENUM(AstcRecompression, Uncompressed, Bc1, Bc3);
ENUM(FramePacingMode, Target_Auto, Target_30, Target_60, Target_90, Target_120);
+1 -5
View File
@@ -119,7 +119,6 @@ struct System::Impl {
is_multicore = Settings::values.use_multi_core.GetValue();
extended_memory_layout = Settings::values.memory_layout_mode.GetValue() != Settings::MemoryLayout::Memory_4Gb;
unified_memory = Settings::values.use_unified_memory.GetValue();
core_timing.SetMulticore(is_multicore);
core_timing.Initialize([&system]() { system.RegisterHostThread(); });
@@ -147,8 +146,7 @@ struct System::Impl {
!device_memory.has_value() ||
is_multicore != Settings::values.use_multi_core.GetValue() ||
extended_memory_layout != (Settings::values.memory_layout_mode.GetValue() !=
Settings::MemoryLayout::Memory_4Gb) ||
unified_memory != Settings::values.use_unified_memory.GetValue();
Settings::MemoryLayout::Memory_4Gb);
if (!must_reinitialize) {
return;
@@ -159,7 +157,6 @@ struct System::Impl {
is_multicore = Settings::values.use_multi_core.GetValue();
extended_memory_layout =
Settings::values.memory_layout_mode.GetValue() != Settings::MemoryLayout::Memory_4Gb;
unified_memory = Settings::values.use_unified_memory.GetValue();
Initialize(system);
}
@@ -506,7 +503,6 @@ struct System::Impl {
std::atomic_bool is_powered_on{};
bool is_multicore : 1 = false;
bool extended_memory_layout : 1 = false;
bool unified_memory : 1 = false;
bool exit_locked : 1 = false;
bool exit_requested : 1 = false;
bool nvdec_active : 1 = false;
+1 -13
View File
@@ -1,6 +1,3 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
@@ -15,18 +12,9 @@ constexpr size_t VirtualReserveSize = 1ULL << 38;
constexpr size_t VirtualReserveSize = 1ULL << 39;
#endif
namespace {
size_t ApplicationPoolOffset() {
using Init = Kernel::Board::Nintendo::Nx::KSystemControl::Init;
const size_t dram_size = Init::GetIntendedMemorySize();
const size_t application_pool_size = Init::GetApplicationPoolSize();
return dram_size > application_pool_size ? dram_size - application_pool_size : 0;
}
}
DeviceMemory::DeviceMemory()
: buffer{Kernel::Board::Nintendo::Nx::KSystemControl::Init::GetIntendedMemorySize(),
VirtualReserveSize, ApplicationPoolOffset()} {}
VirtualReserveSize} {}
DeviceMemory::~DeviceMemory() = default;
-35
View File
@@ -20,8 +20,6 @@
#include "common/scratch_buffer.h"
#include "common/virtual_buffer.h"
struct AHardwareBuffer;
namespace Core {
constexpr size_t DEVICE_PAGEBITS = 12ULL;
@@ -97,34 +95,6 @@ public:
ApplyOpOnPAddr(address, buffer, operation);
}
u8* GetPhysicalBase() noexcept {
return reinterpret_cast<u8*>(physical_base);
}
const u8* GetPhysicalBase() const noexcept {
return reinterpret_cast<const u8*>(physical_base);
}
size_t GetPhysicalSize() const noexcept {
return physical_size;
}
std::span<AHardwareBuffer* const> GetBackingHardwareBuffers() const noexcept {
return ahb_windows;
}
size_t GetBackingHardwareBufferWindowSize() const noexcept {
return ahb_window_size;
}
size_t GetBackingHardwareBufferBase() const noexcept {
return ahb_base;
}
bool IsBackingShared() const noexcept {
return backing_is_shared;
}
PAddr GetPhysicalRawAddressFromDAddr(DAddr address) const {
PAddr subbits = PAddr(address & page_mask);
auto paddr = tracked_entries[(address >> page_bits)].compressed_physical_ptr;
@@ -201,11 +171,6 @@ private:
std::unique_ptr<DeviceMemoryManagerAllocator<Traits>> impl;
const uintptr_t physical_base;
const size_t physical_size;
const std::span<AHardwareBuffer* const> ahb_windows;
const size_t ahb_window_size;
const size_t ahb_base;
const bool backing_is_shared;
DeviceInterface* device_inter;
struct TrackedEntry {
-5
View File
@@ -171,11 +171,6 @@ struct DeviceMemoryManagerAllocator {
template <typename Traits>
DeviceMemoryManager<Traits>::DeviceMemoryManager(const DeviceMemory& device_memory_)
: physical_base{uintptr_t(device_memory_.buffer.BackingBasePointer())}
, physical_size{device_memory_.buffer.BackingSize()}
, ahb_windows{device_memory_.buffer.BackingHardwareBuffers()}
, ahb_window_size{device_memory_.buffer.BackingHardwareBufferWindowSize()}
, ahb_base{device_memory_.buffer.BackingHardwareBufferBase()}
, backing_is_shared{device_memory_.buffer.IsBackingShared()}
, device_inter{nullptr}
, compressed_device_addr(1ULL << ((Settings::values.memory_layout_mode.GetValue() == Settings::MemoryLayout::Memory_4Gb ? physical_min_bits : physical_max_bits) - Memory::YUZU_PAGEBITS))
, tracked_entries(device_as_size >> Memory::YUZU_PAGEBITS)
+1 -6
View File
@@ -105,12 +105,7 @@ void LoopProcess(Core::System& system) {
auto server_manager = std::make_unique<ServerManager>(system);
auto module = std::make_shared<Module>(system);
server_manager->RegisterNamedService("nvdrv", std::make_shared<NVDRV>(system, module, "nvdrv"));
const auto NvdrvInterfaceFactoryForApplets = [&, module] {
return std::make_shared<NVDRV>(system, module, "nvdrv:a");
};
server_manager->RegisterNamedService("nvdrv:a", NvdrvInterfaceFactoryForApplets);
server_manager->RegisterNamedService("nvdrv:a", std::make_shared<NVDRV>(system, module, "nvdrv:a"));
server_manager->RegisterNamedService("nvdrv:s", std::make_shared<NVDRV>(system, module, "nvdrv:s"));
server_manager->RegisterNamedService("nvdrv:t", std::make_shared<NVDRV>(system, module, "nvdrv:t"));
server_manager->RegisterNamedService("nvmemp", std::make_shared<NVMEMP>(system));
+1 -6
View File
@@ -588,12 +588,7 @@ void LoopProcess(Core::System& system) {
auto server_manager = std::make_unique<ServerManager>(system);
auto ro = std::make_shared<RoContext>();
const auto RoInterfaceFactoryForUser = [&, ro] {
return std::make_shared<RoInterface>(system, "ldr:ro", ro, NrrKind::User);
};
server_manager->RegisterNamedService("ldr:ro", std::move(RoInterfaceFactoryForUser));
server_manager->RegisterNamedService("ldr:ro", std::make_shared<RoInterface>(system, "ldr:ro", ro, NrrKind::User));
server_manager->RegisterNamedService("ro:1", std::make_shared<RoInterface>(system, "ro:1", ro, NrrKind::JitPlugin));
server_manager->RegisterNamedService("ro:dmnt", std::make_shared<IDebugMonitorInterface>(system));
ServerManager::RunServer(std::move(server_manager));
+3 -2
View File
@@ -227,8 +227,6 @@ std::unique_ptr<TranslationMap> InitializeTranslations(QObject* parent) {
tr("Preserves GPU-modified data by reading it back before uploading.\nSome games require this to render certain effects properly."));
INSERT(Settings, use_asynchronous_shaders, tr("Enable asynchronous shader compilation"),
tr("May reduce shader stutter."));
INSERT(Settings, use_unified_memory, tr("Enable unified memory access"),
tr("Lets the GPU write buffer readbacks directly into guest memory."));
INSERT(Settings, gpu_clock, tr("GPU Clocks"),
tr("Makes the game believe GPU work finishes faster than it does, so it stops lowering "
"resolution and render distance to fit the Switch's clocks."));
@@ -524,6 +522,9 @@ std::unique_ptr<ComboboxTranslationMap> ComboboxEnumeration(QObject* parent) {
PAIR(AnisotropyMode, X4, tr("4x")),
PAIR(AnisotropyMode, X8, tr("8x")),
PAIR(AnisotropyMode, X16, tr("16x")),
PAIR(AnisotropyMode, X32, tr("32x")),
PAIR(AnisotropyMode, X64, tr("64x")),
PAIR(AnisotropyMode, None, tr("None")),
}});
translations->insert(
{Settings::EnumMetadata<Settings::Language>::Index(),
@@ -335,7 +335,7 @@ void SetupDenormControl(const Profile& profile, const IR::Program& program, Emit
if (info.uses_fp32_denorms_flush && info.uses_fp32_denorms_preserve) {
LOG_DEBUG(Shader_SPIRV, "Fp32 denorm flush and preserve on the same shader");
} else if (info.uses_fp32_denorms_flush) {
if (profile.support_fp32_denorm_flush && !profile.has_broken_fp32_denorm_flush) {
if (profile.support_fp32_denorm_flush) {
ctx.AddCapability(spv::Capability::DenormFlushToZero);
ctx.AddExecutionMode(main_func, spv::ExecutionMode::DenormFlushToZero, 32U);
} else {
@@ -13,7 +13,6 @@
namespace Shader::Backend::SPIRV {
namespace {
Id SharedPointer(EmitContext& ctx, Id offset, u32 index_offset = 0) {
offset = ctx.BoundSharedOffset(offset, 4 + index_offset * 4);
const Id shift_id{ctx.Const(2U)};
Id index{ctx.OpShiftRightArithmetic(ctx.U32[1], offset, shift_id)};
if (index_offset > 0) {
@@ -161,8 +160,7 @@ Id EmitSharedAtomicExchange32(EmitContext& ctx, Id offset, Id value) {
Id EmitSharedAtomicExchange64(EmitContext& ctx, Id offset, Id value) {
if (ctx.profile.support_shared_int64_atomics && ctx.uses_explicit_workgroup_layout) {
const Id shift_id{ctx.Const(3U)};
const Id index{
ctx.OpShiftRightArithmetic(ctx.U32[1], ctx.BoundSharedOffset(offset, 8), shift_id)};
const Id index{ctx.OpShiftRightArithmetic(ctx.U32[1], offset, shift_id)};
const Id pointer{
ctx.OpAccessChain(ctx.shared_u64, ctx.shared_memory_u64, ctx.u32_zero_value, index)};
const auto [scope, semantics]{AtomicArgs(ctx)};
@@ -31,7 +31,6 @@ std::pair<Id, Id> ExtractArgs(EmitContext& ctx, Id offset, u32 mask, u32 count)
} // Anonymous namespace
Id EmitLoadSharedU8(EmitContext& ctx, Id offset) {
offset = ctx.BoundSharedOffset(offset, 1);
if (ctx.uses_explicit_workgroup_layout) {
const Id pointer{
ctx.OpAccessChain(ctx.shared_u8, ctx.shared_memory_u8, ctx.u32_zero_value, offset)};
@@ -43,7 +42,6 @@ Id EmitLoadSharedU8(EmitContext& ctx, Id offset) {
}
Id EmitLoadSharedS8(EmitContext& ctx, Id offset) {
offset = ctx.BoundSharedOffset(offset, 1);
if (ctx.uses_explicit_workgroup_layout) {
const Id pointer{
ctx.OpAccessChain(ctx.shared_u8, ctx.shared_memory_u8, ctx.u32_zero_value, offset)};
@@ -55,7 +53,6 @@ Id EmitLoadSharedS8(EmitContext& ctx, Id offset) {
}
Id EmitLoadSharedU16(EmitContext& ctx, Id offset) {
offset = ctx.BoundSharedOffset(offset, 2);
if (ctx.uses_explicit_workgroup_layout) {
const Id pointer{Pointer(ctx, ctx.shared_u16, ctx.shared_memory_u16, offset, 1)};
return ctx.OpUConvert(ctx.U32[1], ctx.OpLoad(ctx.U16, pointer));
@@ -66,7 +63,6 @@ Id EmitLoadSharedU16(EmitContext& ctx, Id offset) {
}
Id EmitLoadSharedS16(EmitContext& ctx, Id offset) {
offset = ctx.BoundSharedOffset(offset, 2);
if (ctx.uses_explicit_workgroup_layout) {
const Id pointer{Pointer(ctx, ctx.shared_u16, ctx.shared_memory_u16, offset, 1)};
return ctx.OpSConvert(ctx.U32[1], ctx.OpLoad(ctx.U16, pointer));
@@ -77,7 +73,6 @@ Id EmitLoadSharedS16(EmitContext& ctx, Id offset) {
}
Id EmitLoadSharedU32(EmitContext& ctx, Id offset) {
offset = ctx.BoundSharedOffset(offset, 4);
if (ctx.uses_explicit_workgroup_layout) {
const Id pointer{Pointer(ctx, ctx.shared_u32, ctx.shared_memory_u32, offset, 2)};
return ctx.OpLoad(ctx.U32[1], pointer);
@@ -87,7 +82,6 @@ Id EmitLoadSharedU32(EmitContext& ctx, Id offset) {
}
Id EmitLoadSharedU64(EmitContext& ctx, Id offset) {
offset = ctx.BoundSharedOffset(offset, 8);
if (ctx.uses_explicit_workgroup_layout) {
const Id pointer{Pointer(ctx, ctx.shared_u32x2, ctx.shared_memory_u32x2, offset, 3)};
return ctx.OpLoad(ctx.U32[2], pointer);
@@ -103,7 +97,6 @@ Id EmitLoadSharedU64(EmitContext& ctx, Id offset) {
}
Id EmitLoadSharedU128(EmitContext& ctx, Id offset) {
offset = ctx.BoundSharedOffset(offset, 16);
if (ctx.uses_explicit_workgroup_layout) {
const Id pointer{Pointer(ctx, ctx.shared_u32x4, ctx.shared_memory_u32x4, offset, 4)};
return ctx.OpLoad(ctx.U32[4], pointer);
@@ -120,7 +113,6 @@ Id EmitLoadSharedU128(EmitContext& ctx, Id offset) {
}
void EmitWriteSharedU8(EmitContext& ctx, Id offset, Id value) {
offset = ctx.BoundSharedOffset(offset, 1);
if (ctx.uses_explicit_workgroup_layout) {
const Id pointer{
ctx.OpAccessChain(ctx.shared_u8, ctx.shared_memory_u8, ctx.u32_zero_value, offset)};
@@ -131,7 +123,6 @@ void EmitWriteSharedU8(EmitContext& ctx, Id offset, Id value) {
}
void EmitWriteSharedU16(EmitContext& ctx, Id offset, Id value) {
offset = ctx.BoundSharedOffset(offset, 2);
if (ctx.uses_explicit_workgroup_layout) {
const Id pointer{Pointer(ctx, ctx.shared_u16, ctx.shared_memory_u16, offset, 1)};
ctx.OpStore(pointer, ctx.OpUConvert(ctx.U16, value));
@@ -141,7 +132,6 @@ void EmitWriteSharedU16(EmitContext& ctx, Id offset, Id value) {
}
void EmitWriteSharedU32(EmitContext& ctx, Id offset, Id value) {
offset = ctx.BoundSharedOffset(offset, 4);
Id pointer{};
if (ctx.uses_explicit_workgroup_layout) {
pointer = Pointer(ctx, ctx.shared_u32, ctx.shared_memory_u32, offset, 2);
@@ -154,7 +144,6 @@ void EmitWriteSharedU32(EmitContext& ctx, Id offset, Id value) {
}
void EmitWriteSharedU64(EmitContext& ctx, Id offset, Id value) {
offset = ctx.BoundSharedOffset(offset, 8);
if (ctx.uses_explicit_workgroup_layout) {
const Id pointer{Pointer(ctx, ctx.shared_u32x2, ctx.shared_memory_u32x2, offset, 3)};
ctx.OpStore(pointer, value);
@@ -170,7 +159,6 @@ void EmitWriteSharedU64(EmitContext& ctx, Id offset, Id value) {
}
void EmitWriteSharedU128(EmitContext& ctx, Id offset, Id value) {
offset = ctx.BoundSharedOffset(offset, 16);
if (ctx.uses_explicit_workgroup_layout) {
const Id pointer{Pointer(ctx, ctx.shared_u32x4, ctx.shared_memory_u32x4, offset, 4)};
ctx.OpStore(pointer, value);
@@ -600,16 +600,6 @@ void EmitContext::DefineLocalMemory(const IR::Program& program) {
}
}
Id EmitContext::BoundSharedOffset(Id offset, u32 access_bytes) {
if (shared_memory_declared_bytes == 0) {
return offset;
}
const u32 last_valid{shared_memory_declared_bytes > access_bytes
? shared_memory_declared_bytes - access_bytes
: 0U};
return OpUMin(U32[1], offset, Const(last_valid));
}
void EmitContext::DefineSharedMemory(const IR::Program& program) {
uses_explicit_workgroup_layout =
profile.support_explicit_workgroup_layout &&
@@ -618,15 +608,8 @@ void EmitContext::DefineSharedMemory(const IR::Program& program) {
if (program.shared_memory_size == 0) {
return;
}
const u32 device_limit{profile.max_shared_memory_size};
const u32 shared_memory_size{device_limit != 0 && program.shared_memory_size > device_limit
? device_limit
: program.shared_memory_size};
if (shared_memory_size != program.shared_memory_size) {
shared_memory_declared_bytes = shared_memory_size;
}
const auto make{[&](Id element_type, u32 element_size) {
const u32 num_elements{Common::DivCeil(shared_memory_size, element_size)};
const u32 num_elements{Common::DivCeil(program.shared_memory_size, element_size)};
const Id array_type{TypeArray(element_type, Const(num_elements))};
Decorate(array_type, spv::Decoration::ArrayStride, element_size);
@@ -661,7 +644,7 @@ void EmitContext::DefineSharedMemory(const IR::Program& program) {
std::tie(shared_memory_u32x4, shared_u32x4, std::ignore) = make(U32[4], 16);
return;
}
const u32 num_elements{Common::DivCeil(shared_memory_size, 4U)};
const u32 num_elements{Common::DivCeil(program.shared_memory_size, 4U)};
const Id type{TypeArray(U32[1], Const(num_elements))};
shared_memory_u32_type = TypePointer(spv::StorageClass::Workgroup, type);
@@ -312,8 +312,6 @@ public:
Id local_memory{};
bool uses_explicit_workgroup_layout{};
u32 shared_memory_declared_bytes{};
[[nodiscard]] Id BoundSharedOffset(Id offset, u32 access_bytes);
Id shared_memory_u8{};
Id shared_memory_u16{};
Id shared_memory_u32{};
-5
View File
@@ -86,8 +86,6 @@ struct Profile {
bool has_broken_signed_operations{};
/// Float controls break when fp16 is enabled
bool has_broken_fp16_float_controls{};
/// Declaring fp32 denorm flush to zero miscompiles on some drivers
bool has_broken_fp32_denorm_flush{};
/// Dynamic vec4 indexing is broken on some OpenGL drivers
bool has_gl_component_indexing_bug{};
/// The precise type qualifier is broken in the fragment stage of some drivers
@@ -103,9 +101,6 @@ struct Profile {
u32 gl_max_compute_smem_size{};
/// Largest workgroup shared memory allocation the device accepts, 0 when unconstrained
u32 max_shared_memory_size{};
/// Maxwell and earlier nVidia architectures have broken robust support
bool has_broken_robust{};
+44 -345
View File
@@ -571,11 +571,7 @@ void BufferCache<P>::AccumulateFlushes() {
template <class P>
bool BufferCache<P>::ShouldWaitAsyncFlushes() const noexcept {
if (async_buffers.empty()) {
return false;
}
return async_buffers.front().has_value() ||
!pending_downloads.front().unified_copies.empty();
return (!async_buffers.empty() && async_buffers.front().has_value());
}
template <class P>
@@ -583,7 +579,6 @@ void BufferCache<P>::CommitAsyncFlushesHigh() {
AccumulateFlushes();
if (committed_gpu_modified_ranges.empty()) {
pending_downloads.emplace_back();
async_buffers.emplace_back(std::optional<Async_Buffer>{});
return;
}
@@ -643,84 +638,27 @@ void BufferCache<P>::CommitAsyncFlushesHigh() {
}
committed_gpu_modified_ranges.clear();
if (downloads.empty()) {
pending_downloads.emplace_back();
async_buffers.emplace_back(std::optional<Async_Buffer>{});
return;
}
struct QueuedUnifiedCopy {
u64 window;
BufferId buffer_id;
boost::container::small_vector<BufferCopy, 16> copies;
};
AsyncDownloadBatch batch;
boost::container::small_vector<std::pair<BufferCopy, BufferId>, 16> staging_downloads;
boost::container::small_vector<QueuedUnifiedCopy, 4> unified_copy_queue;
boost::container::small_vector<u64, 4> window_ids;
UnifiedWindowGroups groups;
u64 staging_size_bytes = 0;
for (auto& [copy, buffer_id] : downloads) {
Buffer& buffer = slot_buffers[buffer_id];
const DAddr orig_device_addr = buffer.CpuAddr() + copy.src_offset;
bool unified = false;
if constexpr (USE_UNIFIED_MEMORY) {
if (runtime.HasUnifiedMemory()) {
window_ids.clear();
groups.clear();
unified = ResolveUnifiedWindows(orig_device_addr, copy.src_offset, copy.size,
window_ids, groups);
}
}
BufferCopy record{copy};
record.src_offset = static_cast<size_t>(orig_device_addr);
if (unified) {
async_downloads.Add(orig_device_addr, copy.size);
buffer.MarkUsage(copy.src_offset, copy.size);
for (size_t i = 0; i < window_ids.size(); ++i) {
unified_copy_queue.push_back(
QueuedUnifiedCopy{window_ids[i], buffer_id, std::move(groups[i])});
}
batch.unified_copies.push_back(record);
continue;
}
copy.dst_offset = staging_size_bytes;
constexpr u64 align = 64ULL;
staging_size_bytes += (copy.size + align - 1) & ~(align - 1ULL);
staging_downloads.push_back({copy, buffer_id});
}
std::optional<Async_Buffer> download_staging;
if (!staging_downloads.empty()) {
download_staging = runtime.DownloadStagingBuffer(staging_size_bytes, true);
}
auto download_staging = runtime.DownloadStagingBuffer(total_size_bytes, true);
boost::container::small_vector<BufferCopy, 4> normalized_copies;
runtime.PreCopyBarrier();
for (auto& [copy, buffer_id] : staging_downloads) {
copy.dst_offset += download_staging->offset;
for (auto& [copy, buffer_id] : downloads) {
copy.dst_offset += download_staging.offset;
const std::array copies{copy};
BufferCopy second_copy{copy};
Buffer& buffer = slot_buffers[buffer_id];
BufferCopy record{copy};
record.src_offset = static_cast<size_t>(buffer.CpuAddr()) + copy.src_offset;
const DAddr orig_device_addr = static_cast<DAddr>(record.src_offset);
second_copy.src_offset = static_cast<size_t>(buffer.CpuAddr()) + copy.src_offset;
const DAddr orig_device_addr = static_cast<DAddr>(second_copy.src_offset);
async_downloads.Add(orig_device_addr, copy.size);
buffer.MarkUsage(copy.src_offset, copy.size);
runtime.CopyBuffer(download_staging->buffer, buffer, copies, false);
batch.staging_copies.push_back(record);
}
if constexpr (USE_UNIFIED_MEMORY) {
for (const auto& queued : unified_copy_queue) {
const std::span<const BufferCopy> group_span(queued.copies.data(),
queued.copies.size());
runtime.CopyToUnifiedMemory(queued.window, slot_buffers[queued.buffer_id], group_span);
}
if (!unified_copy_queue.empty()) {
runtime.FlushUnifiedMemoryCopies();
runtime.UnifiedMemoryHostBarrier();
}
runtime.CopyBuffer(download_staging.buffer, buffer, copies, false);
normalized_copies.push_back(second_copy);
}
runtime.PostCopyBarrier();
pending_downloads.emplace_back(std::move(batch));
async_buffers.emplace_back(std::move(download_staging));
pending_downloads.emplace_back(std::move(normalized_copies));
async_buffers.emplace_back(download_staging);
}
template <class P>
@@ -735,49 +673,32 @@ void BufferCache<P>::PopAsyncFlushes() {
template <class P>
void BufferCache<P>::PopAsyncBuffers() {
struct Writeback {
DAddr addr;
const u8* src;
u64 size;
};
boost::container::small_vector<Writeback, 8> writebacks;
{
std::scoped_lock lock{mutex};
if (async_buffers.empty()) {
return;
}
auto& batch = pending_downloads.front();
auto& async_buffer = async_buffers.front();
if (async_buffer.has_value()) {
const u8* base = async_buffer->mapped_span.data();
const size_t base_offset = async_buffer->offset;
for (const auto& copy : batch.staging_copies) {
const DAddr device_addr = static_cast<DAddr>(copy.src_offset);
const u64 dst_offset = copy.dst_offset - base_offset;
const u8* read_mapped_memory = base + dst_offset;
async_downloads.ForEachInRange(
device_addr, copy.size, [&](DAddr start, DAddr end, s32) {
writebacks.push_back(
{start, &read_mapped_memory[start - device_addr], end - start});
});
async_downloads.Subtract(device_addr, copy.size, [&](DAddr start, DAddr end) {
gpu_modified_ranges.Subtract(start, end - start);
});
}
async_buffers_death_ring.emplace_back(*async_buffer);
}
for (const auto& copy : batch.unified_copies) {
const DAddr device_addr = static_cast<DAddr>(copy.src_offset);
async_downloads.Subtract(device_addr, copy.size, [&](DAddr start, DAddr end) {
gpu_modified_ranges.Subtract(start, end - start);
});
}
if (async_buffers.empty()) {
return;
}
if (!async_buffers.front().has_value()) {
async_buffers.pop_front();
pending_downloads.pop_front();
return;
}
for (const auto& wb : writebacks) {
device_memory.WriteBlockUnsafe(wb.addr, wb.src, wb.size);
auto& downloads = pending_downloads.front();
auto& async_buffer = async_buffers.front();
u8* base = async_buffer->mapped_span.data();
const size_t base_offset = async_buffer->offset;
for (const auto& copy : downloads) {
const DAddr device_addr = static_cast<DAddr>(copy.src_offset);
const u64 dst_offset = copy.dst_offset - base_offset;
const u8* read_mapped_memory = base + dst_offset;
async_downloads.ForEachInRange(device_addr, copy.size, [&](DAddr start, DAddr end, s32) {
device_memory.WriteBlockUnsafe(start, &read_mapped_memory[start - device_addr],
end - start);
});
async_downloads.Subtract(device_addr, copy.size, [&](DAddr start, DAddr end) {
gpu_modified_ranges.Subtract(start, end - start);
});
}
async_buffers_death_ring.emplace_back(*async_buffer);
async_buffers.pop_front();
pending_downloads.pop_front();
}
template <class P>
@@ -821,20 +742,6 @@ void BufferCache<P>::BindHostIndexBuffer() {
const u32 size = channel_state->index_buffer.size;
const auto& draw_state = maxwell3d->draw_manager.draw_state;
if (draw_state.inline_index_draw_indexes.empty()) {
if constexpr (USE_UNIFIED_MEMORY && !HAS_FULL_INDEX_AND_PRIMITIVE_SUPPORT) {
const auto window =
TryResolveUnifiedRange(channel_state->index_buffer.device_addr, size);
if (window && runtime.IsUnifiedIndexRange(draw_state.topology,
draw_state.index_buffer.format,
window->offset)) {
runtime.BindIndexBuffer(draw_state.topology, draw_state.index_buffer.format,
draw_state.index_buffer.first,
draw_state.index_buffer.count,
runtime.UnifiedWindowBuffer(window->window),
static_cast<u32>(window->offset), size);
return;
}
}
SynchronizeBuffer(buffer, channel_state->index_buffer.device_addr, size);
} else {
if constexpr (USE_MEMORY_MAPS_FOR_UPLOADS) {
@@ -888,7 +795,6 @@ void BufferCache<P>::UpdateVertexBufferSlot(u32 index, const Binding& binding) {
enabled_vertex_buffers_mask |= (1u << index);
} else {
enabled_vertex_buffers_mask &= ~(1u << index);
unified_vertex_buffers_mask &= ~(1u << index);
}
}
@@ -921,39 +827,15 @@ void BufferCache<P>::BindHostVertexBuffers() {
const Binding& binding = VertexBufferSlot(index);
Buffer& buffer = slot_buffers[binding.buffer_id];
TouchBuffer(buffer, binding.buffer_id);
bool needs_bind = flags[Dirty::VertexBuffer0 + index];
u32 unified_window = NO_UNIFIED_WINDOW;
u64 unified_offset = 0;
if constexpr (USE_UNIFIED_MEMORY) {
const bool was_unified = ((unified_vertex_buffers_mask >> index) & 1) != 0;
if (needs_bind || was_unified) {
const auto window = TryResolveUnifiedRange(binding.device_addr, binding.size);
if (window) {
unified_window = static_cast<u32>(window->window);
unified_offset = window->offset;
}
}
if (was_unified && unified_window == NO_UNIFIED_WINDOW) {
needs_bind = true;
}
}
if (unified_window == NO_UNIFIED_WINDOW) {
SynchronizeBuffer(buffer, binding.device_addr, binding.size);
}
if (!needs_bind) {
SynchronizeBuffer(buffer, binding.device_addr, binding.size);
if (!flags[Dirty::VertexBuffer0 + index]) {
flush_bindings();
continue;
}
flags[Dirty::VertexBuffer0 + index] = false;
const u32 stride = maxwell3d->regs.vertex_streams[index].stride;
u32 offset = static_cast<u32>(unified_offset);
if (unified_window == NO_UNIFIED_WINDOW) {
offset = buffer.Offset(binding.device_addr);
buffer.MarkUsage(offset, binding.size);
unified_vertex_buffers_mask &= ~(1u << index);
} else {
unified_vertex_buffers_mask |= 1u << index;
}
const u32 offset = buffer.Offset(binding.device_addr);
buffer.MarkUsage(offset, binding.size);
if (!bindings.buffers.empty() && index != last_index + 1) {
flush_bindings();
}
@@ -964,7 +846,6 @@ void BufferCache<P>::BindHostVertexBuffers() {
bindings.offsets.push_back(offset);
bindings.sizes.push_back(binding.size);
bindings.strides.push_back(stride);
bindings.unified_windows.push_back(unified_window);
last_index = index;
}
flush_bindings();
@@ -972,31 +853,12 @@ void BufferCache<P>::BindHostVertexBuffers() {
HostBindings<typename P::Buffer> host_bindings;
bool any_valid{false};
auto& flags = maxwell3d->dirty.flags;
std::array<u32, NUM_VERTEX_BUFFERS> unified_windows;
std::array<u64, NUM_VERTEX_BUFFERS> unified_offsets{};
unified_windows.fill(NO_UNIFIED_WINDOW);
for (u32 index = 0; index < NUM_VERTEX_BUFFERS; ++index) {
const Binding& binding = channel_state->vertex_buffers[index];
Buffer& buffer = slot_buffers[binding.buffer_id];
TouchBuffer(buffer, binding.buffer_id);
bool needs_bind = flags[Dirty::VertexBuffer0 + index];
if constexpr (USE_UNIFIED_MEMORY) {
const bool was_unified = ((unified_vertex_buffers_mask >> index) & 1) != 0;
if (needs_bind || was_unified) {
const auto window = TryResolveUnifiedRange(binding.device_addr, binding.size);
if (window) {
unified_windows[index] = static_cast<u32>(window->window);
unified_offsets[index] = window->offset;
}
}
if (was_unified && unified_windows[index] == NO_UNIFIED_WINDOW) {
needs_bind = true;
}
}
if (unified_windows[index] == NO_UNIFIED_WINDOW) {
SynchronizeBuffer(buffer, binding.device_addr, binding.size);
}
if (!needs_bind) {
SynchronizeBuffer(buffer, binding.device_addr, binding.size);
if (!flags[Dirty::VertexBuffer0 + index]) {
continue;
}
flags[Dirty::VertexBuffer0 + index] = false;
@@ -1015,20 +877,13 @@ void BufferCache<P>::BindHostVertexBuffers() {
Buffer& buffer = slot_buffers[binding.buffer_id];
const u32 stride = maxwell3d->regs.vertex_streams[index].stride;
u32 offset = static_cast<u32>(unified_offsets[index]);
if (unified_windows[index] == NO_UNIFIED_WINDOW) {
offset = buffer.Offset(binding.device_addr);
buffer.MarkUsage(offset, binding.size);
unified_vertex_buffers_mask &= ~(1u << index);
} else {
unified_vertex_buffers_mask |= 1u << index;
}
const u32 offset = buffer.Offset(binding.device_addr);
buffer.MarkUsage(offset, binding.size);
host_bindings.buffers.push_back(&buffer);
host_bindings.offsets.push_back(offset);
host_bindings.sizes.push_back(binding.size);
host_bindings.strides.push_back(stride);
host_bindings.unified_windows.push_back(unified_windows[index]);
}
runtime.BindVertexBuffers(host_bindings);
}
@@ -1151,25 +1006,11 @@ void BufferCache<P>::BindHostGraphicsStorageBuffers(size_t stage) {
Buffer& buffer = slot_buffers[binding.buffer_id];
TouchBuffer(buffer, binding.buffer_id);
const u32 size = binding.size;
const bool is_written = ((channel_state->written_storage_buffers[stage] >> index) & 1) != 0;
if constexpr (USE_UNIFIED_MEMORY) {
const auto window = TryResolveUnifiedRange(binding.device_addr, size);
if (window && runtime.IsUnifiedStorageRange(size, window->offset)) {
if (is_written) {
memory_tracker.MarkRegionAsCpuModified(binding.device_addr, size);
}
runtime.BindStorageBuffer(runtime.UnifiedWindowBuffer(window->window),
runtime.UnifiedWindowAddress(window->window),
static_cast<u32>(window->offset), size, is_written);
return;
}
}
SynchronizeBuffer(buffer, binding.device_addr, size);
const u32 offset = buffer.Offset(binding.device_addr);
buffer.MarkUsage(offset, size);
const bool is_written = ((channel_state->written_storage_buffers[stage] >> index) & 1) != 0;
if (is_written) {
MarkWrittenBuffer(binding.buffer_id, binding.device_addr, size);
@@ -1858,142 +1699,6 @@ void BufferCache<P>::ImmediateUploadMemory([[maybe_unused]] Buffer& buffer,
}
}
template <class P>
bool BufferCache<P>::ResolveUnifiedWindows(
[[maybe_unused]] DAddr device_addr, [[maybe_unused]] u64 buffer_offset,
[[maybe_unused]] u64 size, [[maybe_unused]] boost::container::small_vector<u64, 4>& window_ids,
[[maybe_unused]] UnifiedWindowGroups& groups) {
if constexpr (USE_UNIFIED_MEMORY) {
const u8* const physical_base = device_memory.GetPhysicalBase();
const u64 unified_base = runtime.UnifiedMemoryBase();
const u64 unified_size = runtime.UnifiedMemorySize();
const u64 window_size = runtime.UnifiedMemoryWindowSize();
if (window_size == 0) {
return false;
}
const auto group_for = [&](u64 window) -> boost::container::small_vector<BufferCopy, 16>& {
for (size_t i = 0; i < window_ids.size(); ++i) {
if (window_ids[i] == window) {
return groups[i];
}
}
window_ids.push_back(window);
groups.emplace_back();
return groups.back();
};
u64 downloaded = 0;
while (downloaded < size) {
const DAddr page_addr = device_addr + downloaded;
const u8* const ptr = device_memory.GetPointer<u8>(page_addr);
if (ptr == nullptr) {
return false;
}
const u64 page_offset = page_addr & Core::DEVICE_PAGEMASK;
u64 chunk = (std::min)(size - downloaded,
static_cast<u64>(Core::DEVICE_PAGESIZE) - page_offset);
const u64 phys_offset = static_cast<u64>(ptr - physical_base);
if (phys_offset < unified_base || phys_offset - unified_base + chunk > unified_size) {
return false;
}
const u64 relative = phys_offset - unified_base;
const u64 window = relative / window_size;
const u64 local_offset = relative % window_size;
chunk = (std::min)(chunk, window_size - local_offset);
auto& group = group_for(window);
if (!group.empty()) {
BufferCopy& last = group.back();
if (last.src_offset + last.size == buffer_offset + downloaded &&
last.dst_offset + last.size == local_offset) {
last.size += chunk;
downloaded += chunk;
continue;
}
}
group.push_back(BufferCopy{
.src_offset = buffer_offset + downloaded,
.dst_offset = local_offset,
.size = chunk,
});
downloaded += chunk;
}
return true;
} else {
return false;
}
}
template <class P>
std::optional<typename BufferCache<P>::UnifiedWindowRange>
BufferCache<P>::TryResolveUnifiedRange([[maybe_unused]] DAddr device_addr,
[[maybe_unused]] u64 size) {
if constexpr (USE_UNIFIED_MEMORY) {
if (size == 0 || !runtime.IsUnifiedMemoryBindable()) {
return std::nullopt;
}
const u64 window_size = runtime.UnifiedMemoryWindowSize();
if (window_size == 0) {
return std::nullopt;
}
const u8* const first = device_memory.GetSpan(device_addr, size);
if (first == nullptr) {
return std::nullopt;
}
const u64 phys_offset = static_cast<u64>(first - device_memory.GetPhysicalBase());
const u64 unified_base = runtime.UnifiedMemoryBase();
if (phys_offset < unified_base) {
return std::nullopt;
}
const u64 relative = phys_offset - unified_base;
const u64 unified_size = runtime.UnifiedMemorySize();
if (relative >= unified_size || unified_size - relative < size) {
return std::nullopt;
}
const u64 local_offset = relative % window_size;
if (window_size - local_offset < size) {
return std::nullopt;
}
if (memory_tracker.IsRegionGpuModified(device_addr, size) ||
IsRegionGpuModified(device_addr, size)) {
return std::nullopt;
}
return UnifiedWindowRange{
.window = static_cast<size_t>(relative / window_size),
.offset = local_offset,
};
} else {
return std::nullopt;
}
}
template <class P>
bool BufferCache<P>::TryUnifiedDownloadMemory([[maybe_unused]] Buffer& buffer,
[[maybe_unused]] std::span<BufferCopy> copies) {
if constexpr (USE_UNIFIED_MEMORY) {
boost::container::small_vector<u64, 4> window_ids;
UnifiedWindowGroups groups;
for (const BufferCopy& copy : copies) {
if (!ResolveUnifiedWindows(buffer.CpuAddr() + copy.src_offset, copy.src_offset,
copy.size, window_ids, groups)) {
return false;
}
}
for (const BufferCopy& copy : copies) {
buffer.MarkUsage(copy.src_offset, copy.size);
}
runtime.PreCopyBarrier();
for (size_t i = 0; i < window_ids.size(); ++i) {
const std::span<const BufferCopy> group_span(groups[i].data(), groups[i].size());
runtime.CopyToUnifiedMemory(window_ids[i], buffer, group_span);
}
runtime.FlushUnifiedMemoryCopies();
runtime.UnifiedMemoryHostBarrier();
runtime.Finish();
return true;
} else {
return false;
}
}
template <class P>
void BufferCache<P>::MappedUploadMemory([[maybe_unused]] Buffer& buffer,
[[maybe_unused]] u64 total_size_bytes,
@@ -2097,12 +1802,6 @@ void BufferCache<P>::DownloadBufferMemory(Buffer& buffer, DAddr device_addr, u64
}
if constexpr (USE_MEMORY_MAPS) {
if constexpr (USE_UNIFIED_MEMORY) {
if (runtime.HasUnifiedMemory() &&
TryUnifiedDownloadMemory(buffer, std::span(copies.data(), copies.size()))) {
return;
}
}
auto download_staging = runtime.DownloadStagingBuffer(total_size_bytes);
const u8* const mapped_memory = download_staging.mapped_span.data();
const std::span<BufferCopy> copies_span(copies.data(), copies.data() + copies.size());
@@ -10,10 +10,8 @@
#include <array>
#include <bit>
#include <functional>
#include <limits>
#include <memory>
#include <mutex>
#include <optional>
#include <numeric>
#include <span>
#include <vector>
@@ -97,15 +95,12 @@ static constexpr Binding NULL_BINDING{
.buffer_id = NULL_BUFFER_ID,
};
static constexpr u32 NO_UNIFIED_WINDOW = (std::numeric_limits<u32>::max)();
template <typename Buffer>
struct HostBindings {
boost::container::static_vector<Buffer*, NUM_VERTEX_BUFFERS> buffers;
boost::container::static_vector<u64, NUM_VERTEX_BUFFERS> offsets;
boost::container::static_vector<u64, NUM_VERTEX_BUFFERS> sizes;
boost::container::static_vector<u64, NUM_VERTEX_BUFFERS> strides;
boost::container::static_vector<u32, NUM_VERTEX_BUFFERS> unified_windows;
u32 min_index{NUM_VERTEX_BUFFERS};
u32 max_index{0};
};
@@ -185,7 +180,6 @@ class BufferCache : public VideoCommon::ChannelSetupCaches<BufferCacheChannelInf
static constexpr bool USE_MEMORY_MAPS = P::USE_MEMORY_MAPS;
static constexpr bool SEPARATE_IMAGE_BUFFERS_BINDINGS = P::SEPARATE_IMAGE_BUFFER_BINDINGS;
static constexpr bool USE_MEMORY_MAPS_FOR_UPLOADS = P::USE_MEMORY_MAPS_FOR_UPLOADS;
static constexpr bool USE_UNIFIED_MEMORY = P::USE_UNIFIED_MEMORY;
#ifdef YUZU_LEGACY
static constexpr s64 TARGET_THRESHOLD = 3_GiB;
@@ -449,22 +443,6 @@ private:
void MappedUploadMemory(Buffer& buffer, u64 total_size_bytes, std::span<BufferCopy> copies);
bool TryUnifiedDownloadMemory(Buffer& buffer, std::span<BufferCopy> copies);
struct UnifiedWindowRange {
size_t window;
u64 offset;
};
std::optional<UnifiedWindowRange> TryResolveUnifiedRange(DAddr device_addr, u64 size);
using UnifiedWindowGroups =
boost::container::small_vector<boost::container::small_vector<BufferCopy, 16>, 4>;
bool ResolveUnifiedWindows(DAddr device_addr, u64 buffer_offset, u64 size,
boost::container::small_vector<u64, 4>& window_ids,
UnifiedWindowGroups& groups);
void DownloadBufferMemory(Buffer& buffer_id);
void DownloadBufferMemory(Buffer& buffer_id, DAddr device_addr, u64 size);
@@ -509,7 +487,6 @@ private:
u32 last_index_count = 0;
u32 enabled_vertex_buffers_mask = 0;
u32 unified_vertex_buffers_mask = 0;
u64 vertex_buffers_serial = 0;
std::array<Binding, 32> v_buffer{};
@@ -521,14 +498,9 @@ private:
std::deque<Common::RangeSet<DAddr>> committed_gpu_modified_ranges;
// Async Buffers
struct AsyncDownloadBatch {
boost::container::small_vector<BufferCopy, 4> staging_copies;
boost::container::small_vector<BufferCopy, 4> unified_copies;
};
Common::OverlapRangeSet<DAddr> async_downloads;
std::deque<std::optional<Async_Buffer>> async_buffers;
std::deque<AsyncDownloadBatch> pending_downloads;
std::deque<boost::container::small_vector<BufferCopy, 4>> pending_downloads;
std::optional<Async_Buffer> current_buffer;
std::deque<Async_Buffer> async_buffers_death_ring;
@@ -7,7 +7,6 @@ layout(push_constant) uniform constants {
vec2 scale;
vec2 size;
vec2 resize_factor;
vec2 crop_offset;
float edge_sharpness;
};
layout(location = 0) out highp vec2 texcoord;
@@ -16,5 +15,5 @@ void main() {
float x = float((gl_VertexIndex & 1) << 2);
float y = float((gl_VertexIndex & 2) << 1);
gl_Position = vec4(x - 1.0f, y - 1.0f, 0.0, 1.0f) * vec4(sign(resize_factor), 1.f, 1.f);
texcoord = crop_offset + vec2(x, y) * abs(resize_factor) * 0.5;
texcoord = vec2(x, y) * abs(resize_factor) * 0.5;
}
@@ -14,7 +14,6 @@ layout(push_constant) uniform constants {
vec2 scale;
vec2 size;
vec2 resize_factor;
vec2 crop_offset;
float edge_sharpness;
};
layout(set = 0, binding = 0) uniform sampler2D sampler0;
@@ -13,7 +13,6 @@
layout( push_constant ) uniform constants {
vec4 ViewportInfo[1];
vec2 ResizeFactor;
vec2 CropOffset;
float EdgeSharpness;
};
layout(set = 0, binding = 0) uniform sampler2D ps0;
@@ -261,7 +261,6 @@ struct BufferCacheParams {
// TODO: Investigate why OpenGL seems to perform worse with persistently mapped buffer uploads
static constexpr bool USE_MEMORY_MAPS_FOR_UPLOADS = false;
static constexpr bool USE_UNIFIED_MEMORY = false;
};
using BufferCache = VideoCommon::BufferCache<BufferCacheParams>;
@@ -231,14 +231,12 @@ ShaderCache::ShaderCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
.has_broken_unsigned_image_offsets = true,
.has_broken_signed_operations = true,
.has_broken_fp16_float_controls = false,
.has_broken_fp32_denorm_flush = false,
.has_gl_component_indexing_bug = device.HasComponentIndexingBug(),
.has_gl_precise_bug = device.HasPreciseBug(),
.has_gl_cbuf_ftou_bug = device.HasCbufFtouBug(),
.has_gl_bool_ref_bug = device.HasBoolRefBug(),
.ignore_nan_fp_comparisons = true,
.gl_max_compute_smem_size = device.GetMaxComputeSharedMemorySize(),
.max_shared_memory_size = device.GetMaxComputeSharedMemorySize(),
.min_ssbo_alignment = device.GetShaderStorageBufferAlignment(),
// Use the host limit, but never more than the guest can produce. Maxwell exposes 8 clip
// distances and the SPIR-V output array is sized for at most 8, so clamping here keeps a
@@ -1281,7 +1281,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceColorPipeline(const BlitImagePipelineKe
.subpass = 0,
.basePipelineHandle = VK_NULL_HANDLE,
.basePipelineIndex = 0,
}));
}, device.StaticPipelineCache()));
return *blit_color_pipelines.back();
}
@@ -1313,7 +1313,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceDepthStencilPipeline(const BlitImagePip
.subpass = 0,
.basePipelineHandle = VK_NULL_HANDLE,
.basePipelineIndex = 0,
}));
}, device.StaticPipelineCache()));
return *blit_depth_stencil_pipelines.back();
}
@@ -1366,7 +1366,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceClearColorPipeline(const BlitImagePipel
.subpass = 0,
.basePipelineHandle = VK_NULL_HANDLE,
.basePipelineIndex = 0,
}));
}, device.StaticPipelineCache()));
return *clear_color_pipelines.back();
}
@@ -1422,7 +1422,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceClearStencilPipeline(
.subpass = 0,
.basePipelineHandle = VK_NULL_HANDLE,
.basePipelineIndex = 0,
}));
}, device.StaticPipelineCache()));
return *clear_stencil_pipelines.back();
}
@@ -1465,7 +1465,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceBlitColorMSAAPipeline(const BlitMSAAPip
.subpass = 0,
.basePipelineHandle = VK_NULL_HANDLE,
.basePipelineIndex = 0,
}));
}, device.StaticPipelineCache()));
return *blit_msaa_color_pipelines.back();
}
@@ -1584,7 +1584,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceResolveDepthStencilPipeline(VkRenderPas
.subpass = 0,
.basePipelineHandle = VK_NULL_HANDLE,
.basePipelineIndex = 0,
}));
}, device.StaticPipelineCache()));
return *pipelines.back();
}
@@ -1697,7 +1697,7 @@ VkPipeline BlitImageHelper::FindOrEmplaceMSAACopyPipeline(const MSAACopyPipeline
.subpass = 0,
.basePipelineHandle = VK_NULL_HANDLE,
.basePipelineIndex = 0,
}));
}, device.StaticPipelineCache()));
return *msaa_copy_pipelines.back();
}
@@ -1817,7 +1817,7 @@ void BlitImageHelper::ConvertPipelineEx(vk::Pipeline& pipeline, VkRenderPass ren
.subpass = 0,
.basePipelineHandle = VK_NULL_HANDLE,
.basePipelineIndex = 0,
});
}, device.StaticPipelineCache());
}
void BlitImageHelper::ConvertPipelineColorTargetEx(vk::Pipeline& pipeline, VkRenderPass renderpass,
@@ -1860,7 +1860,7 @@ void BlitImageHelper::ConvertPipeline(vk::Pipeline& pipeline, VkRenderPass rende
.subpass = 0,
.basePipelineHandle = VK_NULL_HANDLE,
.basePipelineIndex = 0,
});
}, device.StaticPipelineCache());
}
} // namespace Vulkan
@@ -370,9 +370,20 @@ inline void PushImageDescriptors(TextureCache& texture_cache,
const VkImageView null_image_view{texture_cache.GetImageView(VideoCommon::NULL_IMAGE_VIEW_ID).Handle(desc.type)};
if (null_image_view != VK_NULL_HANDLE) vk_image_view = null_image_view;
}
Sampler& sampler{texture_cache.GetSampler(sampler_id)};
guest_descriptor_queue.AddSampledImage(vk_image_view,
sampler.HandleFor(image_view, desc.is_depth));
const Sampler& sampler{texture_cache.GetSampler(sampler_id)};
const bool use_fallback_sampler{sampler.HasAddedAnisotropy() &&
!image_view.SupportsAnisotropy()};
VkSampler vk_sampler{use_fallback_sampler ? sampler.HandleWithDefaultAnisotropy()
: sampler.Handle()};
if (sampler.HasLinearFiltering() &&
VideoCore::Surface::IsPixelFormatInteger(image_view.format)) {
vk_sampler = sampler.HandleWithNearestFilter();
}
if (desc.is_depth && sampler.HasDepthComparison() &&
!image_view.SupportsDepthComparison()) {
vk_sampler = sampler.HandleWithoutDepthComparison();
}
guest_descriptor_queue.AddSampledImage(vk_image_view, vk_sampler);
const bool element_rescaled{texture_cache.IsRescaling(image_view)};
is_rescaled |= element_rescaled;
}
+14 -18
View File
@@ -1,8 +1,6 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
#include <algorithm>
#include "common/common_types.h"
#include "common/div_ceil.h"
#include "common/settings.h"
@@ -19,7 +17,7 @@
namespace Vulkan {
using PushConstants = std::array<u32, 4 + 2 + 2 + 1>;
using PushConstants = std::array<u32, 4 + 2 + 1>;
SGSR::SGSR(const Device& device, MemoryAllocator& memory_allocator, size_t image_count, VkExtent2D extent, bool edge_dir)
: m_memory_allocator{memory_allocator}
@@ -102,28 +100,26 @@ VkImageView SGSR::Draw(const Device& device, Scheduler& scheduler, size_t image_
const f32 input_image_width = f32(input_image_extent.width);
const f32 input_image_height = f32(input_image_extent.height);
const f32 crop_width = (crop_rect.right - crop_rect.left) * input_image_width;
const f32 crop_height = (crop_rect.bottom - crop_rect.top) * input_image_height;
static constexpr f32 EDGE_SHARPNESS_MAX = 2.0f;
const f32 edge_sharpness =
EDGE_SHARPNESS_MAX - f32(Settings::values.fsr_sharpening_slider.GetValue()) / 200.0f;
const f32 viewport_width = (crop_rect.right - crop_rect.left) * input_image_width;
const f32 viewport_height = (crop_rect.bottom - crop_rect.top) * input_image_height;
// expected [0, 2]
const f32 sharpening = f32(Settings::values.fsr_sharpening_slider.GetValue()) / 100.0f;
// p = (tex * viewport) / input = [0,n] (normalized texcoords)
// p * input = [0,1024], [0,768]
// layout( push_constant ) uniform constants {
// highp vec4 ViewportInfo[1];
// highp vec2 ResizeFactor;
// highp vec2 CropOffset;
// highp float EdgeSharpness;
// };
PushConstants viewport_con{};
viewport_con[0] = std::bit_cast<u32>(1.f / input_image_width);
viewport_con[1] = std::bit_cast<u32>(1.f / input_image_height);
viewport_con[2] = std::bit_cast<u32>(input_image_width);
viewport_con[3] = std::bit_cast<u32>(input_image_height);
viewport_con[4] = std::bit_cast<u32>(crop_width / input_image_width);
viewport_con[5] = std::bit_cast<u32>(crop_height / input_image_height);
viewport_con[6] = std::bit_cast<u32>((std::min)(crop_rect.left, crop_rect.right));
viewport_con[7] = std::bit_cast<u32>((std::min)(crop_rect.top, crop_rect.bottom));
viewport_con[8] = std::bit_cast<u32>(edge_sharpness);
viewport_con[0] = std::bit_cast<u32>(std::abs(1.f / viewport_width));
viewport_con[1] = std::bit_cast<u32>(std::abs(1.f / viewport_height));
viewport_con[2] = std::bit_cast<u32>(std::abs(viewport_width));
viewport_con[3] = std::bit_cast<u32>(std::abs(viewport_height));
viewport_con[4] = std::bit_cast<u32>(viewport_width / input_image_width);
viewport_con[5] = std::bit_cast<u32>(viewport_height / input_image_height);
viewport_con[6] = std::bit_cast<u32>(sharpening);
UploadImages(device, scheduler);
UpdateDescriptorSets(device, source_image_view, image_index);
@@ -521,7 +521,7 @@ static vk::Pipeline CreateWrappedPipelineImpl(
.subpass = 0,
.basePipelineHandle = 0,
.basePipelineIndex = 0,
});
}, device.StaticPipelineCache());
}
vk::Pipeline CreateWrappedPipeline(const Device& device, vk::RenderPass& renderpass,
@@ -7,9 +7,7 @@
#include <algorithm>
#include <array>
#include <cstring>
#include <limits>
#include <span>
#include <utility>
#include <vector>
#include "video_core/buffer_cache/buffer_cache_base.h"
@@ -34,32 +32,6 @@ VkBufferCopy MakeBufferCopy(const VideoCommon::BufferCopy& copy) {
};
}
constexpr size_t MAX_WINDOW_BARRIER_RANGES = 8;
using WindowRange = std::pair<VkDeviceSize, VkDeviceSize>;
using WindowRanges = boost::container::small_vector<WindowRange, MAX_WINDOW_BARRIER_RANGES>;
void CoalesceWindowRanges(WindowRanges& ranges) {
if (ranges.size() < 2) {
return;
}
std::sort(ranges.begin(), ranges.end());
size_t merged = 0;
for (size_t index = 1; index < ranges.size(); ++index) {
if (ranges[index].first <= ranges[merged].second) {
ranges[merged].second = (std::max)(ranges[merged].second, ranges[index].second);
} else {
ranges[++merged] = ranges[index];
}
}
ranges.resize(merged + 1);
if (ranges.size() > MAX_WINDOW_BARRIER_RANGES) {
const WindowRange bounding{ranges.front().first, ranges.back().second};
ranges.clear();
ranges.push_back(bounding);
}
}
VkIndexType IndexTypeFromNumElements(const Device& device, u32 num_elements) {
if (num_elements <= 0xff && device.IsExtIndexTypeUint8Supported()) {
return VK_INDEX_TYPE_UINT8_EXT;
@@ -70,16 +42,6 @@ VkIndexType IndexTypeFromNumElements(const Device& device, u32 num_elements) {
return VK_INDEX_TYPE_UINT32;
}
u32 GrowIndexCount(u32 current, u32 requested) {
constexpr u32 MinimumIndices = 4096;
constexpr u32 GrowthLimit = (std::numeric_limits<u32>::max)() / 2;
u32 grown = (std::max)(requested, MinimumIndices);
if (current <= GrowthLimit) {
grown = (std::max)(grown, current * 2);
}
return grown;
}
size_t BytesPerIndex(VkIndexType index_type) {
switch (index_type) {
case VK_INDEX_TYPE_UINT8_EXT:
@@ -196,12 +158,13 @@ public:
virtual ~QuadIndexBuffer() = default;
void UpdateBuffer(u32 num_indices_) {
ReleaseRetiredBuffers();
if (num_indices_ <= num_indices) {
return;
}
num_indices = GrowIndexCount(num_indices, num_indices_);
scheduler.Finish();
num_indices = num_indices_;
index_type = IndexTypeFromNumElements(device, num_indices);
const u32 num_quads = GetQuadsNum(num_indices);
@@ -219,12 +182,6 @@ public:
.queueFamilyIndexCount = 0,
.pQueueFamilyIndices = nullptr,
};
if (buffer) {
retired_buffers.push_back(RetiredBuffer{
.buffer = std::move(buffer),
.tick = scheduler.CurrentTick(),
});
}
buffer = memory_allocator.CreateBuffer(buffer_ci, MemoryUsage::DeviceLocal);
if (device.HasDebuggingToolAttached()) {
buffer.SetObjectNameEXT("Quad LUT");
@@ -292,17 +249,6 @@ protected:
virtual void MakeAndUpdateIndices(u8* staging_data, size_t quad_size, u32 quad, u32 first) = 0;
struct RetiredBuffer {
vk::Buffer buffer;
u64 tick;
};
void ReleaseRetiredBuffers() {
std::erase_if(retired_buffers, [this](const RetiredBuffer& entry) {
return scheduler.IsFree(entry.tick);
});
}
const Device& device;
MemoryAllocator& memory_allocator;
Scheduler& scheduler;
@@ -310,7 +256,6 @@ protected:
vk::Buffer buffer{};
MemoryCommit memory_commit{};
std::vector<RetiredBuffer> retired_buffers;
VkIndexType index_type{};
u32 num_indices = 0;
};
@@ -420,141 +365,6 @@ BufferCacheRuntime::BufferCacheRuntime(const Device& device_, MemoryAllocator& m
scheduler_, staging_pool_);
}
void BufferCacheRuntime::TryEnableUnifiedMemory(void* base, size_t size,
std::span<AHardwareBuffer* const> hardware_buffers,
size_t hardware_buffer_window,
size_t hardware_buffer_base) {
unified_memory = std::make_unique<HostMemoryImport>(
device, base, size, hardware_buffers, hardware_buffer_window, hardware_buffer_base);
if (!unified_memory->IsValid()) {
unified_memory.reset();
}
}
void BufferCacheRuntime::CopyToUnifiedMemory(
size_t window_index, VkBuffer src_buffer,
std::span<const VideoCommon::BufferCopy> copies) {
if (!unified_memory || src_buffer == VK_NULL_HANDLE || copies.empty() ||
window_index >= unified_memory->GetWindowCount() ||
unified_memory->GetWindowBuffer(window_index) == VK_NULL_HANDLE) {
return;
}
PendingUnifiedCopy& pending = pending_unified_copies.emplace_back();
pending.window = window_index;
pending.buffer = src_buffer;
pending.copies.resize(copies.size());
std::ranges::transform(copies, pending.copies.begin(), MakeBufferCopy);
}
void BufferCacheRuntime::FlushUnifiedMemoryCopies() {
if (pending_unified_copies.empty()) {
return;
}
struct UnifiedCopyCommand {
VkBuffer buffer;
boost::container::small_vector<VkBufferCopy, 8> copies;
};
std::stable_sort(pending_unified_copies.begin(), pending_unified_copies.end(),
[](const PendingUnifiedCopy& lhs, const PendingUnifiedCopy& rhs) {
return lhs.window < rhs.window;
});
const bool foreign = unified_memory->NeedsForeignOwnershipTransfer();
const u32 queue_family = device.GetGraphicsFamily();
size_t group_begin = 0;
while (group_begin < pending_unified_copies.size()) {
const size_t window = pending_unified_copies[group_begin].window;
size_t group_end = group_begin;
while (group_end < pending_unified_copies.size() &&
pending_unified_copies[group_end].window == window) {
++group_end;
}
const VkBuffer window_buffer = unified_memory->GetWindowBuffer(window);
WindowRanges ranges;
for (size_t index = group_begin; index < group_end; ++index) {
for (const VkBufferCopy& copy : pending_unified_copies[index].copies) {
ranges.emplace_back(copy.dstOffset, copy.dstOffset + copy.size);
}
}
CoalesceWindowRanges(ranges);
boost::container::small_vector<VkBufferMemoryBarrier, MAX_WINDOW_BARRIER_RANGES> acquire;
boost::container::small_vector<VkBufferMemoryBarrier, MAX_WINDOW_BARRIER_RANGES> release;
if (foreign) {
for (const WindowRange& range : ranges) {
acquire.push_back(VkBufferMemoryBarrier{
.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER,
.pNext = nullptr,
.srcAccessMask = 0,
.dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_FOREIGN_EXT,
.dstQueueFamilyIndex = queue_family,
.buffer = window_buffer,
.offset = range.first,
.size = range.second - range.first,
});
release.push_back(VkBufferMemoryBarrier{
.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER,
.pNext = nullptr,
.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
.dstAccessMask = 0,
.srcQueueFamilyIndex = queue_family,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_FOREIGN_EXT,
.buffer = window_buffer,
.offset = range.first,
.size = range.second - range.first,
});
}
}
boost::container::small_vector<UnifiedCopyCommand, 4> commands;
commands.reserve(group_end - group_begin);
for (size_t index = group_begin; index < group_end; ++index) {
PendingUnifiedCopy& pending = pending_unified_copies[index];
commands.push_back(UnifiedCopyCommand{pending.buffer, std::move(pending.copies)});
}
scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([window_buffer, acquire = std::move(acquire), release = std::move(release),
commands = std::move(commands)](vk::CommandBuffer cmdbuf) {
if (!acquire.empty()) {
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT,
VK_PIPELINE_STAGE_TRANSFER_BIT, 0, {},
VideoCommon::FixSmallVectorADL(acquire), {});
}
for (const UnifiedCopyCommand& command : commands) {
cmdbuf.CopyBuffer(command.buffer, window_buffer,
VideoCommon::FixSmallVectorADL(command.copies));
}
if (!release.empty()) {
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, 0, {},
VideoCommon::FixSmallVectorADL(release), {});
}
});
group_begin = group_end;
}
pending_unified_copies.clear();
}
void BufferCacheRuntime::UnifiedMemoryHostBarrier() {
static constexpr VkMemoryBarrier HOST_BARRIER{
.sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
.pNext = nullptr,
.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
.dstAccessMask = VK_ACCESS_HOST_READ_BIT,
};
scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([](vk::CommandBuffer cmdbuf) {
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_HOST_BIT, 0,
HOST_BARRIER);
});
}
StagingBufferRef BufferCacheRuntime::UploadStagingBuffer(size_t size) {
return staging_pool.Request(size, MemoryUsage::Upload);
}
@@ -592,7 +402,6 @@ u32 BufferCacheRuntime::GetStorageBufferAlignment() const {
}
void BufferCacheRuntime::TickFrame(Common::SlotVector<Buffer>& slot_buffers) noexcept {
FlushUnifiedMemoryCopies();
for (auto it = slot_buffers.begin(); it != slot_buffers.end(); it++) {
if (scheduler.IsFree(it->LastUsageTick())) {
it->ResetUsageTracking();
@@ -727,25 +536,6 @@ void BufferCacheRuntime::ClearBuffer(VkBuffer dest_buffer, u32 offset, size_t si
});
}
bool BufferCacheRuntime::IsUnifiedIndexRange(PrimitiveTopology topology, IndexFormat index_format,
u64 offset) const {
const VkIndexType vk_index_type = MaxwellToVK::IndexFormat(index_format);
const bool needs_uint8_pass =
vk_index_type == VK_INDEX_TYPE_UINT8_EXT && !device.IsExtIndexTypeUint8Supported();
if (topology == PrimitiveTopology::Quads || topology == PrimitiveTopology::QuadStrip ||
needs_uint8_pass) {
return (offset % device.GetStorageBufferAlignment()) == 0;
}
switch (vk_index_type) {
case VK_INDEX_TYPE_UINT32:
return (offset % 4) == 0;
case VK_INDEX_TYPE_UINT16:
return (offset % 2) == 0;
default:
return true;
}
}
void BufferCacheRuntime::BindIndexBuffer(PrimitiveTopology topology, IndexFormat index_format,
u32 base_vertex, u32 num_indices, VkBuffer buffer,
u32 offset, [[maybe_unused]] u32 size) {
@@ -821,11 +611,6 @@ void BufferCacheRuntime::BindVertexBuffer(u32 index, VkBuffer buffer, u32 offset
void BufferCacheRuntime::BindVertexBuffers(VideoCommon::HostBindings<Buffer>& bindings) {
boost::container::static_vector<VkBuffer, VideoCommon::NUM_VERTEX_BUFFERS> buffer_handles(bindings.buffers.size());
for (u32 i = 0; i < bindings.buffers.size(); ++i) {
if (i < bindings.unified_windows.size() &&
bindings.unified_windows[i] != VideoCommon::NO_UNIFIED_WINDOW) {
buffer_handles[i] = unified_memory->GetWindowBuffer(bindings.unified_windows[i]);
continue;
}
auto handle = bindings.buffers[i]->Handle();
if (handle == VK_NULL_HANDLE) {
bindings.offsets[i] = 0;
@@ -7,10 +7,6 @@
#pragma once
#include <limits>
#include <memory>
#include <span>
#include <boost/container/small_vector.hpp>
#include "video_core/buffer_cache/buffer_cache_base.h"
#include "video_core/buffer_cache/memory_tracker_base.h"
@@ -101,33 +97,6 @@ public:
void TickFrame(Common::SlotVector<Buffer>& slot_buffers) noexcept;
void TryEnableUnifiedMemory(void* base, size_t size,
std::span<AHardwareBuffer* const> hardware_buffers,
size_t hardware_buffer_window, size_t hardware_buffer_base);
[[nodiscard]] bool HasUnifiedMemory() const noexcept {
return unified_memory != nullptr && unified_memory->IsValid();
}
[[nodiscard]] u64 UnifiedMemorySize() const noexcept {
return unified_memory ? unified_memory->GetSize() : 0;
}
[[nodiscard]] u64 UnifiedMemoryBase() const noexcept {
return unified_memory ? unified_memory->GetBaseOffset() : 0;
}
[[nodiscard]] u64 UnifiedMemoryWindowSize() const noexcept {
return unified_memory ? unified_memory->GetWindowSize() : 0;
}
void CopyToUnifiedMemory(size_t window_index, VkBuffer src_buffer,
std::span<const VideoCommon::BufferCopy> copies);
void FlushUnifiedMemoryCopies();
void UnifiedMemoryHostBarrier();
u64 CurrentTick();
u64 KnownGpuTick();
@@ -195,32 +164,6 @@ public:
BindBuffer(buffer, offset, size);
}
void BindStorageBuffer(VkBuffer buffer, VkDeviceAddress address, u32 offset, u32 size,
[[maybe_unused]] bool is_written) {
guest_descriptor_queue.AddBuffer(buffer, address, offset, size);
}
[[nodiscard]] bool IsUnifiedMemoryBindable() const noexcept {
return unified_memory != nullptr && unified_memory->IsValid() &&
unified_memory->IsBindable();
}
[[nodiscard]] VkBuffer UnifiedWindowBuffer(size_t index) const noexcept {
return unified_memory->GetWindowBuffer(index);
}
[[nodiscard]] VkDeviceAddress UnifiedWindowAddress(size_t index) const noexcept {
return unified_memory->GetWindowAddress(index);
}
[[nodiscard]] bool IsUnifiedStorageRange(u32 size, u64 offset) const {
return size <= device.GetMaxStorageBufferRange() &&
(offset % device.GetStorageBufferAlignment()) == 0;
}
[[nodiscard]] bool IsUnifiedIndexRange(PrimitiveTopology topology, IndexFormat index_format,
u64 offset) const;
void BindTextureBuffer(Buffer& buffer, u32 offset, u32 size,
VideoCore::Surface::PixelFormat format) {
guest_descriptor_queue.AddTexelBuffer(buffer.View(offset, size, format),
@@ -237,12 +180,6 @@ public:
}
private:
struct PendingUnifiedCopy {
size_t window;
VkBuffer buffer;
boost::container::small_vector<VkBufferCopy, 8> copies;
};
void BindBuffer(const Buffer& buffer, u32 offset, u32 size) {
const VkBuffer handle = buffer.Handle();
if (handle == VK_NULL_HANDLE) {
@@ -267,8 +204,6 @@ private:
std::shared_ptr<QuadStripIndexBuffer> quad_strip_index_buffer;
vk::Buffer null_buffer;
std::unique_ptr<HostMemoryImport> unified_memory;
boost::container::small_vector<PendingUnifiedCopy, 8> pending_unified_copies;
std::unique_ptr<Uint8Pass> uint8_pass;
QuadIndexedPass quad_index_pass;
@@ -291,7 +226,6 @@ struct BufferCacheParams {
static constexpr bool USE_MEMORY_MAPS = true;
static constexpr bool SEPARATE_IMAGE_BUFFER_BINDINGS = false;
static constexpr bool USE_MEMORY_MAPS_FOR_UPLOADS = true;
static constexpr bool USE_UNIFIED_MEMORY = true;
};
using BufferCache = VideoCommon::BufferCache<BufferCacheParams>;
@@ -268,7 +268,7 @@ ComputePass::ComputePass(const Device& device_, Scheduler& scheduler, Descriptor
.layout = *layout,
.basePipelineHandle = {},
.basePipelineIndex = 0,
});
}, device.StaticPipelineCache());
}
ComputePass::~ComputePass() = default;
@@ -63,6 +63,8 @@ using VideoCommon::GenericEnvironment;
using VideoCommon::GraphicsEnvironment;
constexpr u32 CACHE_VERSION = 18;
constexpr size_t VULKAN_CACHE_FLUSH_PIPELINES = 128;
constexpr size_t VULKAN_CACHE_FLUSH_MIN_SECONDS = 30;
constexpr std::array<char, 8> VULKAN_CACHE_MAGIC_NUMBER{'y', 'u', 'z', 'u', 'v', 'k', 'c', 'h'};
template <typename Container>
@@ -444,10 +446,8 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
.has_broken_unsigned_image_offsets = false,
.has_broken_signed_operations = false,
.has_broken_fp16_float_controls = driver_id == VK_DRIVER_ID_NVIDIA_PROPRIETARY,
.has_broken_fp32_denorm_flush = driver_id == VK_DRIVER_ID_QUALCOMM_PROPRIETARY,
.ignore_nan_fp_comparisons = false,
.has_broken_spirv_subgroup_mask_vector_extract_dynamic = false,
.max_shared_memory_size = device.GetMaxComputeSharedMemorySize(),
.has_broken_robust =
device.IsNvidia() && device.GetNvidiaArch() <= NvidiaArchitecture::Arch_Pascal,
.min_ssbo_alignment = device.GetStorageBufferAlignment(),
@@ -701,6 +701,10 @@ void PipelineCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading
if (use_vulkan_pipeline_cache) {
SerializeVulkanPipelineCache(vulkan_pipeline_cache_filename, vulkan_pipeline_cache,
CACHE_VERSION);
size_t size = 0;
vulkan_pipeline_cache.Read(&size, nullptr);
last_cache_size.store(size, std::memory_order_relaxed);
last_flush = std::chrono::steady_clock::now();
}
if (state.statistics) {
@@ -708,6 +712,35 @@ void PipelineCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading
}
}
void PipelineCache::QueueVulkanPipelineCacheFlush() {
if (!use_vulkan_pipeline_cache || vulkan_pipeline_cache_filename.empty()) {
return;
}
if (++pipelines_since_flush < VULKAN_CACHE_FLUSH_PIPELINES) {
return;
}
const auto now = std::chrono::steady_clock::now();
const auto megabytes = last_cache_size.load(std::memory_order_relaxed) / (1024 * 1024);
const std::chrono::seconds interval{
std::max<size_t>(VULKAN_CACHE_FLUSH_MIN_SECONDS, megabytes)};
if (last_flush.time_since_epoch().count() != 0 && now - last_flush < interval) {
return;
}
if (flush_in_flight.exchange(true, std::memory_order_acq_rel)) {
return;
}
pipelines_since_flush = 0;
last_flush = now;
serialization_thread.QueueWork([this] {
SerializeVulkanPipelineCache(vulkan_pipeline_cache_filename, vulkan_pipeline_cache,
CACHE_VERSION);
size_t size = 0;
vulkan_pipeline_cache.Read(&size, nullptr);
last_cache_size.store(size, std::memory_order_relaxed);
flush_in_flight.store(false, std::memory_order_release);
});
}
GraphicsPipeline* PipelineCache::CurrentGraphicsPipelineSlowPath() {
const auto [pair, is_new]{graphics_cache.try_emplace(graphics_key)};
auto& pipeline{pair->second};
@@ -746,7 +779,7 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline(
std::span<Shader::Environment* const> envs, PipelineStatistics* statistics,
bool build_in_parallel) try {
auto hash = key.Hash();
LOG_INFO(Render_Vulkan, "{:#016x}", hash);
LOG_DEBUG(Render_Vulkan, "{:#016x}", hash);
size_t env_index{0};
std::array<Shader::IR::Program, Maxwell::MaxShaderProgram> programs;
const bool uses_vertex_a{key.unique_hashes[0] != 0};
@@ -882,6 +915,7 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline() {
}
SerializePipeline(key, env_ptrs, pipeline_cache_filename, CACHE_VERSION);
});
QueueVulkanPipelineCacheFlush();
return pipeline;
}
@@ -901,6 +935,7 @@ std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
SerializePipeline(key, std::array<const GenericEnvironment*, 1>{&env_},
pipeline_cache_filename, CACHE_VERSION);
});
QueueVulkanPipelineCacheFlush();
return pipeline;
}
@@ -913,7 +948,7 @@ std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
return nullptr;
}
LOG_INFO(Render_Vulkan, "{:#016x}", hash);
LOG_DEBUG(Render_Vulkan, "{:#016x}", hash);
Shader::Maxwell::Flow::CFG cfg{env, pools.flow_block, env.StartAddress()};
@@ -923,6 +958,19 @@ std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
}
auto program{TranslateProgram(pools.inst, pools.block, env, cfg, host_info)};
const VkDriverIdKHR driver_id = device.GetDriverID();
const bool needs_shared_mem_clamp =
driver_id == VK_DRIVER_ID_QUALCOMM_PROPRIETARY ||
driver_id == VK_DRIVER_ID_ARM_PROPRIETARY;
const u32 max_shared_memory = device.GetMaxComputeSharedMemorySize();
if (needs_shared_mem_clamp && program.shared_memory_size > max_shared_memory) {
LOG_WARNING(Render_Vulkan,
"Compute shader {:#016x} requests {}KB shared memory but device max is {}KB - clamping",
key.unique_hash,
program.shared_memory_size / 1024,
max_shared_memory / 1024);
program.shared_memory_size = max_shared_memory;
}
const std::vector<u32> code{EmitSPIRV(profile, program)};
device.SaveShader(code);
vk::ShaderModule spv_module{BuildShader(device, code)};
@@ -7,6 +7,8 @@
#pragma once
#include <array>
#include <atomic>
#include <chrono>
#include <cstddef>
#include <filesystem>
#include <memory>
@@ -144,6 +146,8 @@ private:
vk::PipelineCache LoadVulkanPipelineCache(const std::filesystem::path& filename,
u32 expected_cache_version);
void QueueVulkanPipelineCacheFlush();
const Device& device;
Scheduler& scheduler;
DescriptorPool& descriptor_pool;
@@ -171,6 +175,10 @@ private:
std::filesystem::path vulkan_pipeline_cache_filename;
vk::PipelineCache vulkan_pipeline_cache;
size_t pipelines_since_flush{};
std::chrono::steady_clock::time_point last_flush{};
std::atomic<size_t> last_cache_size{};
std::atomic_bool flush_in_flight{};
Common::ThreadWorker workers;
Common::ThreadWorker serialization_thread;
@@ -225,13 +225,6 @@ RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra
fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache, device, scheduler),
wfi_event(device.GetLogical().CreateEvent()) {
scheduler.SetQueryCache(query_cache);
if (Settings::values.use_unified_memory.GetValue() && device_memory.IsBackingShared()) {
buffer_cache_runtime.TryEnableUnifiedMemory(
device_memory.GetPhysicalBase(), device_memory.GetPhysicalSize(),
device_memory.GetBackingHardwareBuffers(),
device_memory.GetBackingHardwareBufferWindowSize(),
device_memory.GetBackingHardwareBufferBase());
}
}
RasterizerVulkan::~RasterizerVulkan() {
@@ -51,7 +51,6 @@ using VideoCore::Surface::BytesPerBlock;
using VideoCore::Surface::HasAlpha;
using VideoCore::Surface::IsPixelFormatASTC;
using VideoCore::Surface::IsPixelFormatInteger;
using VideoCore::Surface::IsPixelFormatSRGB;
using VideoCore::Surface::SurfaceType;
namespace {
@@ -60,17 +59,6 @@ constexpr bool ENABLE_MSAA_RESOLVE_CONSUME = true;
constexpr bool ENABLE_MSAA_COLOR_DISCARD = true;
constexpr bool ENABLE_MSAA_DEPTH_STENCIL_DISCARD = true;
[[nodiscard]] constexpr bool NeedsExplicitBorderColorFormat(VkFormat format) {
switch (format) {
case VK_FORMAT_B4G4R4A4_UNORM_PACK16:
case VK_FORMAT_B5G6R5_UNORM_PACK16:
case VK_FORMAT_B5G5R5A1_UNORM_PACK16:
return true;
default:
return false;
}
}
constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
if (color == std::array<float, 4>{0, 0, 0, 0}) {
return VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK;
@@ -1814,6 +1802,10 @@ bool TextureCacheRuntime::CanReportMemoryUsage() const {
return device.CanReportMemoryUsage();
}
std::optional<size_t> TextureCacheRuntime::GetSamplerHeapBudget() const {
return device.GetSamplerHeapBudget();
}
void TextureCacheRuntime::FlushDeferredClear() {
scheduler.FlushDeferredClear();
}
@@ -2525,23 +2517,9 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI
supports_depth_comparison =
(properties3.optimalTilingFeatures &
VK_FORMAT_FEATURE_2_SAMPLED_IMAGE_DEPTH_COMPARISON_BIT) != 0;
supports_minmax_filter = (properties3.optimalTilingFeatures &
VK_FORMAT_FEATURE_2_SAMPLED_IMAGE_FILTER_MINMAX_BIT) != 0;
} else {
supports_depth_comparison = true;
supports_minmax_filter =
(device->GetPhysical().GetFormatProperties(format_info.format).optimalTilingFeatures &
VK_FORMAT_FEATURE_SAMPLED_IMAGE_FILTER_MINMAX_BIT) != 0;
}
requires_border_color_format = NeedsExplicitBorderColorFormat(format_info.format);
swizzle_mapping = VkComponentMapping{
.r = ComponentSwizzle(swizzle[0]),
.g = ComponentSwizzle(swizzle[1]),
.b = ComponentSwizzle(swizzle[2]),
.a = ComponentSwizzle(swizzle[3]),
};
has_identity_swizzle = swizzle[0] == SwizzleSource::R && swizzle[1] == SwizzleSource::G &&
swizzle[2] == SwizzleSource::B && swizzle[3] == SwizzleSource::A;
const VkImageUsageFlags requested_view_usage = ImageUsageFlags(format_info, format);
const VkImageUsageFlags image_usage = image.UsageFlags();
const VkImageUsageFlags clamped_view_usage = requested_view_usage & image_usage;
@@ -2566,7 +2544,12 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI
.image = image.Handle(),
.viewType = VkImageViewType{},
.format = format_info.format,
.components = swizzle_mapping,
.components{
.r = ComponentSwizzle(swizzle[0]),
.g = ComponentSwizzle(swizzle[1]),
.b = ComponentSwizzle(swizzle[2]),
.a = ComponentSwizzle(swizzle[3]),
},
.subresourceRange = MakeSubresourceRange(aspect_mask, info.range),
};
const auto create = [&](TextureType tex_type, std::optional<u32> num_layers) {
@@ -2735,223 +2718,92 @@ vk::ImageView ImageView::MakeView(VkFormat vk_format, VkImageAspectFlags aspect_
});
}
CustomBorderColorBudget::~CustomBorderColorBudget() {
Release();
}
CustomBorderColorBudget::CustomBorderColorBudget(CustomBorderColorBudget&& rhs) noexcept
: device_ptr{std::exchange(rhs.device_ptr, nullptr)}, held{std::exchange(rhs.held, 0)} {}
CustomBorderColorBudget& CustomBorderColorBudget::operator=(
CustomBorderColorBudget&& rhs) noexcept {
if (this != &rhs) {
Release();
device_ptr = std::exchange(rhs.device_ptr, nullptr);
held = std::exchange(rhs.held, 0);
}
return *this;
}
bool CustomBorderColorBudget::TryAcquire(const Device& device, size_t count) {
if (!device.TryReserveCustomBorderColorSamplers(count)) {
return false;
}
device_ptr = &device;
held += count;
return true;
}
void CustomBorderColorBudget::Release() noexcept {
if (device_ptr != nullptr) {
device_ptr->ReleaseCustomBorderColorSamplers(held);
}
device_ptr = nullptr;
held = 0;
}
Sampler::Sampler(TextureCacheRuntime& runtime, const Tegra::Texture::TSCEntry& tsc) {
const auto& device = runtime.device;
device_ptr = &device;
border_color = tsc.BorderColor();
srgb_border_color = tsc.SrgbBorderColor();
const bool has_custom_border_extension = runtime.device.IsExtCustomBorderColorSupported();
const bool has_format_undefined =
has_custom_border_extension && runtime.device.IsCustomBorderColorWithoutFormatSupported();
const bool has_custom_border_colors =
has_format_undefined && runtime.device.IsCustomBorderColorsSupported();
const auto color = tsc.BorderColor();
const f32 max_anisotropy = std::clamp(tsc.MaxAnisotropy(), 1.0f, 16.0f);
default_anisotropy = static_cast<f32>(1U << tsc.max_anisotropy);
const VkFilter mag_filter{MaxwellToVK::Sampler::Filter(tsc.mag_filter)};
const VkFilter min_filter{MaxwellToVK::Sampler::Filter(tsc.min_filter)};
const VkSamplerMipmapMode mipmap_mode{MaxwellToVK::Sampler::MipmapMode(tsc.mipmap_filter)};
const VkSamplerAddressMode wrap_u{
MaxwellToVK::Sampler::WrapMode(device, tsc.wrap_u, tsc.mag_filter)};
const VkSamplerAddressMode wrap_v{
MaxwellToVK::Sampler::WrapMode(device, tsc.wrap_v, tsc.mag_filter)};
const VkSamplerAddressMode wrap_p{
MaxwellToVK::Sampler::WrapMode(device, tsc.wrap_p, tsc.mag_filter)};
const bool samples_border = wrap_u == VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER ||
wrap_v == VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER ||
wrap_p == VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER;
reduction_mode = MaxwellToVK::SamplerReduction(tsc.reduction_filter);
has_added_anisotropy = max_anisotropy > default_anisotropy;
has_linear_filtering = mag_filter == VK_FILTER_LINEAR || min_filter == VK_FILTER_LINEAR ||
mipmap_mode == VK_SAMPLER_MIPMAP_MODE_LINEAR;
has_depth_comparison = tsc.depth_compare_enabled != 0;
has_minmax_reduction = reduction_mode != VK_SAMPLER_REDUCTION_MODE_WEIGHTED_AVERAGE_EXT;
has_srgb_border_color = tsc.srgb_conversion != 0 && srgb_border_color != border_color;
if (has_minmax_reduction && !device.IsExtSamplerFilterMinmaxSupported()) {
LOG_WARNING(Render_Vulkan, "VK_EXT_sampler_filter_minmax is required");
has_minmax_reduction = false;
}
has_custom_border_colors = samples_border && device.IsCustomBorderColorUsable();
needs_swizzle_mapping = has_custom_border_colors && device.NeedsBorderColorSwizzleMapping();
if (has_custom_border_colors && GPU::Logging::IsActive()) {
GPU::Logging::GPULogger::GetInstance().LogExtensionUsage(
"VK_EXT_custom_border_color", "Sampler::Sampler");
}
if (device.IsExtBorderColorSwizzleSupported() && GPU::Logging::IsActive()) {
GPU::Logging::GPULogger::GetInstance().LogExtensionUsage(
"VK_EXT_border_color_swizzle", "Sampler::Sampler");
}
f32 min_lod = 0.0f;
f32 max_lod = 0.25f;
if (tsc.mipmap_filter != TextureMipmapFilter::None) {
min_lod = tsc.MinLod();
max_lod = tsc.MaxLod();
}
base_ci = VkSamplerCreateInfo{
.sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.magFilter = mag_filter,
.minFilter = min_filter,
.mipmapMode = mipmap_mode,
.addressModeU = wrap_u,
.addressModeV = wrap_v,
.addressModeW = wrap_p,
.mipLodBias = tsc.LodBias(),
.anisotropyEnable = static_cast<VkBool32>(max_anisotropy > 1.0f),
.maxAnisotropy = max_anisotropy,
.compareEnable = static_cast<VkBool32>(tsc.depth_compare_enabled),
.compareOp = MaxwellToVK::Sampler::DepthCompareFunction(tsc.depth_compare_func),
.minLod = min_lod,
.maxLod = max_lod,
.borderColor = VK_BORDER_COLOR_FLOAT_CUSTOM_EXT,
.unnormalizedCoordinates = VK_FALSE,
};
variants.reserve(1);
Emplace(VariantKey{});
}
Sampler::VariantKey Sampler::MakeKey(const ImageView& image_view, bool is_depth) const noexcept {
VariantKey key{};
key.reduce_anisotropy = has_added_anisotropy && !image_view.SupportsAnisotropy();
key.force_nearest = has_linear_filtering && IsPixelFormatInteger(image_view.format);
key.drop_depth_comparison =
is_depth && has_depth_comparison && !image_view.SupportsDepthComparison();
key.drop_reduction = has_minmax_reduction && !image_view.SupportsMinmaxFilter();
key.drop_custom_border = has_custom_border_colors && image_view.RequiresBorderColorFormat();
key.srgb_border = has_srgb_border_color && IsPixelFormatSRGB(image_view.format);
if (needs_swizzle_mapping && !key.drop_custom_border && !image_view.HasIdentitySwizzle()) {
const VkComponentMapping& mapping = image_view.Swizzle();
key.swizzle = {mapping.r, mapping.g, mapping.b, mapping.a};
}
return key;
}
VkSampler Sampler::Find(const VariantKey& key) const noexcept {
const auto it = std::ranges::find(variants, key, &Variant::key);
if (it == variants.end()) {
return VK_NULL_HANDLE;
}
return *it->sampler;
}
VkSampler Sampler::Emplace(VariantKey key) {
bool custom_border = has_custom_border_colors && !key.drop_custom_border;
if (custom_border && !custom_border_color_budget.TryAcquire(*device_ptr, 1)) {
static bool warned_budget = false;
if (!warned_budget) {
warned_budget = true;
}
custom_border = false;
key.drop_custom_border = true;
key.swizzle = {};
if (const VkSampler existing = Find(key); existing != VK_NULL_HANDLE) {
return existing;
}
}
std::array<float, 4> color = border_color;
if (key.srgb_border) {
color = srgb_border_color;
}
const VkSamplerCustomBorderColorCreateInfoEXT border_ci{
.sType = VK_STRUCTURE_TYPE_SAMPLER_CUSTOM_BORDER_COLOR_CREATE_INFO_EXT,
.pNext = nullptr,
.customBorderColor = std::bit_cast<VkClearColorValue>(color),
.format = VK_FORMAT_UNDEFINED,
};
const VkSamplerBorderColorComponentMappingCreateInfoEXT mapping_ci{
.sType = VK_STRUCTURE_TYPE_SAMPLER_BORDER_COLOR_COMPONENT_MAPPING_CREATE_INFO_EXT,
.pNext = &border_ci,
.components{key.swizzle[0], key.swizzle[1], key.swizzle[2], key.swizzle[3]},
.srgb = VK_FALSE,
};
const void* chain = nullptr;
if (custom_border) {
chain = &border_ci;
if (key.HasSwizzle()) {
chain = &mapping_ci;
const void* pnext = nullptr;
if (has_custom_border_colors) {
pnext = &border_ci;
if (GPU::Logging::IsActive()) {
GPU::Logging::GPULogger::GetInstance().LogExtensionUsage(
"VK_EXT_custom_border_color", "Sampler::Sampler");
}
}
if (device.IsExtBorderColorSwizzleSupported() && GPU::Logging::IsActive()) {
GPU::Logging::GPULogger::GetInstance().LogExtensionUsage(
"VK_EXT_border_color_swizzle", "Sampler::Sampler");
}
const VkSamplerReductionModeCreateInfoEXT reduction_ci{
.sType = VK_STRUCTURE_TYPE_SAMPLER_REDUCTION_MODE_CREATE_INFO_EXT,
.pNext = chain,
.reductionMode = reduction_mode,
.pNext = pnext,
.reductionMode = MaxwellToVK::SamplerReduction(tsc.reduction_filter),
};
if (has_minmax_reduction && !key.drop_reduction) {
chain = &reduction_ci;
if (runtime.device.IsExtSamplerFilterMinmaxSupported()) {
pnext = &reduction_ci;
} else if (reduction_ci.reductionMode != VK_SAMPLER_REDUCTION_MODE_WEIGHTED_AVERAGE_EXT) {
LOG_WARNING(Render_Vulkan, "VK_EXT_sampler_filter_minmax is required");
}
// Some games have samplers with garbage. Sanitize them here.
const f32 max_anisotropy = std::clamp(tsc.MaxAnisotropy(), 1.0f, 16.0f);
VkSamplerCreateInfo create_info = base_ci;
create_info.pNext = chain;
if (key.force_nearest) {
create_info.magFilter = VK_FILTER_NEAREST;
create_info.minFilter = VK_FILTER_NEAREST;
create_info.mipmapMode = VK_SAMPLER_MIPMAP_MODE_NEAREST;
create_info.anisotropyEnable = VK_FALSE;
create_info.maxAnisotropy = 1.0f;
} else if (key.reduce_anisotropy) {
create_info.anisotropyEnable = static_cast<VkBool32>(default_anisotropy > 1.0f);
create_info.maxAnisotropy = default_anisotropy;
}
if (key.drop_depth_comparison) {
create_info.compareEnable = VK_FALSE;
}
if (!custom_border) {
create_info.borderColor = ConvertBorderColor(color);
}
variants.push_back(Variant{
.key = key,
.sampler = device_ptr->GetLogical().CreateSampler(create_info),
});
return *variants.back().sampler;
}
const VkFilter mag_filter{MaxwellToVK::Sampler::Filter(tsc.mag_filter)};
const VkFilter min_filter{MaxwellToVK::Sampler::Filter(tsc.min_filter)};
const VkSamplerMipmapMode mipmap_mode{MaxwellToVK::Sampler::MipmapMode(tsc.mipmap_filter)};
const bool has_linear_filtering{mag_filter == VK_FILTER_LINEAR ||
min_filter == VK_FILTER_LINEAR ||
mipmap_mode == VK_SAMPLER_MIPMAP_MODE_LINEAR};
VkSampler Sampler::HandleFor(const ImageView& image_view, bool is_depth) {
VariantKey key = MakeKey(image_view, is_depth);
if (variants.size() >= MAX_VARIANTS) {
key.srgb_border = false;
key.swizzle = {};
const auto create_sampler = [&](const f32 anisotropy, bool force_nearest,
bool disable_compare = false) {
return device.GetLogical().CreateSampler(VkSamplerCreateInfo{
.sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO,
.pNext = pnext,
.flags = 0,
.magFilter = force_nearest ? VK_FILTER_NEAREST : mag_filter,
.minFilter = force_nearest ? VK_FILTER_NEAREST : min_filter,
.mipmapMode = force_nearest ? VK_SAMPLER_MIPMAP_MODE_NEAREST : mipmap_mode,
.addressModeU = MaxwellToVK::Sampler::WrapMode(device, tsc.wrap_u, tsc.mag_filter),
.addressModeV = MaxwellToVK::Sampler::WrapMode(device, tsc.wrap_v, tsc.mag_filter),
.addressModeW = MaxwellToVK::Sampler::WrapMode(device, tsc.wrap_p, tsc.mag_filter),
.mipLodBias = tsc.LodBias(),
.anisotropyEnable =
static_cast<VkBool32>(!force_nearest && anisotropy > 1.0f ? VK_TRUE : VK_FALSE),
.maxAnisotropy = force_nearest ? 1.0f : anisotropy,
.compareEnable = disable_compare ? VK_FALSE
: static_cast<VkBool32>(tsc.depth_compare_enabled),
.compareOp = MaxwellToVK::Sampler::DepthCompareFunction(tsc.depth_compare_func),
.minLod = tsc.mipmap_filter == TextureMipmapFilter::None ? 0.0f : tsc.MinLod(),
.maxLod = tsc.mipmap_filter == TextureMipmapFilter::None ? 0.25f : tsc.MaxLod(),
.borderColor = has_custom_border_colors ? VK_BORDER_COLOR_FLOAT_CUSTOM_EXT
: ConvertBorderColor(color),
.unnormalizedCoordinates = VK_FALSE,
});
};
sampler = create_sampler(max_anisotropy, false);
const f32 max_anisotropy_default = static_cast<f32>(1U << tsc.max_anisotropy);
if (max_anisotropy > max_anisotropy_default) {
sampler_default_anisotropy = create_sampler(max_anisotropy_default, false);
}
if (const VkSampler existing = Find(key); existing != VK_NULL_HANDLE) {
return existing;
if (has_linear_filtering) {
sampler_nearest = create_sampler(1.0f, true);
}
if (tsc.depth_compare_enabled) {
sampler_noncompare = create_sampler(max_anisotropy, false, true);
}
return Emplace(key);
}
Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM_RT> color_buffers,
@@ -68,6 +68,8 @@ public:
bool CanReportMemoryUsage() const;
std::optional<size_t> GetSamplerHeapBudget() const;
bool CanDownloadMsaa(const VideoCommon::ImageInfo& info) const;
[[nodiscard]] VkImage AcquireMsaaScratchImage(const VkImageCreateInfo& image_ci);
@@ -444,22 +446,6 @@ public:
return supports_depth_comparison;
}
[[nodiscard]] bool RequiresBorderColorFormat() const noexcept {
return requires_border_color_format;
}
[[nodiscard]] bool SupportsMinmaxFilter() const noexcept {
return supports_minmax_filter;
}
[[nodiscard]] const VkComponentMapping& Swizzle() const noexcept {
return swizzle_mapping;
}
[[nodiscard]] bool HasIdentitySwizzle() const noexcept {
return has_identity_swizzle;
}
[[nodiscard]] GPUVAddr GpuAddr() const noexcept {
return gpu_addr;
}
@@ -492,91 +478,48 @@ private:
VkSampleCountFlagBits samples = VK_SAMPLE_COUNT_1_BIT;
u32 buffer_size = 0;
VkComponentMapping swizzle_mapping{};
bool supports_depth_comparison = false;
bool requires_border_color_format = false;
bool supports_minmax_filter = false;
bool has_identity_swizzle = true;
};
class ImageAlloc : public VideoCommon::ImageAllocBase {};
class CustomBorderColorBudget {
public:
CustomBorderColorBudget() = default;
~CustomBorderColorBudget();
CustomBorderColorBudget(const CustomBorderColorBudget&) = delete;
CustomBorderColorBudget& operator=(const CustomBorderColorBudget&) = delete;
CustomBorderColorBudget(CustomBorderColorBudget&& rhs) noexcept;
CustomBorderColorBudget& operator=(CustomBorderColorBudget&& rhs) noexcept;
bool TryAcquire(const Device& device, size_t count);
private:
void Release() noexcept;
const Device* device_ptr = nullptr;
size_t held = 0;
};
class Sampler {
public:
explicit Sampler(TextureCacheRuntime&, const Tegra::Texture::TSCEntry&);
[[nodiscard]] VkSampler Handle() const noexcept {
return *variants.front().sampler;
return *sampler;
}
[[nodiscard]] VkSampler HandleFor(const ImageView& image_view, bool is_depth);
[[nodiscard]] VkSampler HandleWithDefaultAnisotropy() const noexcept {
return *sampler_default_anisotropy;
}
[[nodiscard]] bool HasAddedAnisotropy() const noexcept {
return static_cast<bool>(sampler_default_anisotropy);
}
[[nodiscard]] VkSampler HandleWithNearestFilter() const noexcept {
return *sampler_nearest;
}
[[nodiscard]] bool HasLinearFiltering() const noexcept {
return static_cast<bool>(sampler_nearest);
}
[[nodiscard]] VkSampler HandleWithoutDepthComparison() const noexcept {
return *sampler_noncompare;
}
[[nodiscard]] bool HasDepthComparison() const noexcept {
return static_cast<bool>(sampler_noncompare);
}
private:
struct VariantKey {
bool reduce_anisotropy;
bool force_nearest;
bool drop_depth_comparison;
bool drop_reduction;
bool drop_custom_border;
bool srgb_border;
std::array<VkComponentSwizzle, 4> swizzle;
bool operator==(const VariantKey&) const noexcept = default;
[[nodiscard]] bool HasSwizzle() const noexcept {
return swizzle != std::array<VkComponentSwizzle, 4>{};
}
};
struct Variant {
VariantKey key;
vk::Sampler sampler;
};
static constexpr size_t MAX_VARIANTS = 32;
[[nodiscard]] VariantKey MakeKey(const ImageView& image_view, bool is_depth) const noexcept;
[[nodiscard]] VkSampler Find(const VariantKey& key) const noexcept;
VkSampler Emplace(VariantKey key);
CustomBorderColorBudget custom_border_color_budget;
std::vector<Variant> variants;
const Device* device_ptr{nullptr};
VkSamplerCreateInfo base_ci{};
VkSamplerReductionModeEXT reduction_mode{VK_SAMPLER_REDUCTION_MODE_WEIGHTED_AVERAGE_EXT};
std::array<float, 4> border_color{};
std::array<float, 4> srgb_border_color{};
f32 default_anisotropy{1.0f};
bool has_added_anisotropy{};
bool has_linear_filtering{};
bool has_depth_comparison{};
bool has_minmax_reduction{};
bool has_custom_border_colors{};
bool has_srgb_border_color{};
bool needs_swizzle_mapping{};
vk::Sampler sampler;
vk::Sampler sampler_default_anisotropy;
vk::Sampler sampler_nearest;
vk::Sampler sampler_noncompare;
};
struct TextureCacheParams {
@@ -1896,10 +1896,67 @@ SamplerId TextureCache<P>::FindSampler(const TSCEntry& config, bool compute) {
const auto [pair, is_new] = channel_state->samplers.try_emplace(config);
if (is_new) {
pair->second = slot_samplers.insert(runtime, config);
EnforceSamplerBudget();
}
return pair->second;
}
template <class P>
std::optional<size_t> TextureCache<P>::QuerySamplerBudget() const {
if constexpr (requires { runtime.GetSamplerHeapBudget(); }) {
return runtime.GetSamplerHeapBudget();
} else {
return std::nullopt;
}
}
template <class P>
void TextureCache<P>::EnforceSamplerBudget() {
if (auto const budget = QuerySamplerBudget(); budget) {
if (slot_samplers.size() < *budget) {
return;
}
if (!channel_state) {
return;
}
if (last_sampler_gc_frame == frame_tick) {
return;
}
last_sampler_gc_frame = frame_tick;
TrimInactiveSamplers(*budget);
}
}
template <class P>
void TextureCache<P>::TrimInactiveSamplers(size_t budget) {
if (channel_state->samplers.size() > 0) {
constexpr size_t SAMPLER_GC_SLACK = 1024;
ankerl::unordered_dense::set<SamplerId> active_sampler_ids;
for (auto const& e : channel_state->sampler_ids)
active_sampler_ids.insert(e.second);
// Elements in the map must be necesarily valid
size_t removed = 0;
for (auto it = channel_state->samplers.begin(); it != channel_state->samplers.end();) {
const SamplerId sampler_id = it->second;
if (!sampler_id || sampler_id == CORRUPT_ID) {
it = channel_state->samplers.erase(it);
} else if (std::ranges::find(active_sampler_ids, sampler_id) != active_sampler_ids.end()) {
++it;
} else {
slot_samplers.erase(sampler_id);
it = channel_state->samplers.erase(it);
++removed;
if (slot_samplers.size() + SAMPLER_GC_SLACK <= budget) {
break;
}
}
}
if (removed != 0) {
LOG_WARNING(HW_GPU, "Sampler cache exceeded {} entries on this driver; reclaimed {} inactive samplers", budget, removed);
}
}
}
template <class P>
ImageViewId TextureCache<P>::FindColorBuffer(size_t index) {
const auto& regs = maxwell3d->regs;
@@ -420,6 +420,9 @@ private:
void QueueAsyncDecode(Image& image, ImageId image_id);
void TickAsyncDecode();
void EnforceSamplerBudget();
void TrimInactiveSamplers(size_t budget);
std::optional<size_t> QuerySamplerBudget() const;
void QueueAsyncUnswizzle(Image& image, ImageId image_id);
void TickAsyncUnswizzle();
@@ -506,6 +509,7 @@ private:
u64 modification_tick = 0;
u64 frame_tick = 0;
u64 last_sampler_gc_frame = (std::numeric_limits<u64>::max)();
Common::ThreadWorker texture_decode_worker{1, "TextureDecoder", {},
Common::ThreadPlacement::Efficiency};
+49 -22
View File
@@ -1,11 +1,10 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <array>
#include <cmath>
#include "common/cityhash.h"
#include "common/settings.h"
@@ -18,25 +17,53 @@ namespace Tegra::Texture {
namespace {
float SrgbToLinear(u32 value) {
const float encoded = static_cast<float>(value) / 255.0f;
if (encoded <= 0.04045f) {
return encoded / 12.92f;
}
return std::pow((encoded + 0.055f) / 1.055f, 2.4f);
}
[[maybe_unused]] constexpr std::array<float, 256> SRGB_CONVERSION_LUT = {
0.000000f, 0.000000f, 0.000000f, 0.000012f, 0.000021f, 0.000033f, 0.000046f, 0.000062f,
0.000081f, 0.000102f, 0.000125f, 0.000151f, 0.000181f, 0.000214f, 0.000251f, 0.000293f,
0.000338f, 0.000388f, 0.000443f, 0.000503f, 0.000568f, 0.000639f, 0.000715f, 0.000798f,
0.000887f, 0.000983f, 0.001085f, 0.001195f, 0.001312f, 0.001437f, 0.001569f, 0.001710f,
0.001860f, 0.002019f, 0.002186f, 0.002364f, 0.002551f, 0.002748f, 0.002955f, 0.003174f,
0.003403f, 0.003643f, 0.003896f, 0.004160f, 0.004436f, 0.004725f, 0.005028f, 0.005343f,
0.005672f, 0.006015f, 0.006372f, 0.006744f, 0.007130f, 0.007533f, 0.007950f, 0.008384f,
0.008834f, 0.009301f, 0.009785f, 0.010286f, 0.010805f, 0.011342f, 0.011898f, 0.012472f,
0.013066f, 0.013680f, 0.014313f, 0.014967f, 0.015641f, 0.016337f, 0.017054f, 0.017793f,
0.018554f, 0.019337f, 0.020144f, 0.020974f, 0.021828f, 0.022706f, 0.023609f, 0.024536f,
0.025489f, 0.026468f, 0.027473f, 0.028504f, 0.029563f, 0.030649f, 0.031762f, 0.032904f,
0.034074f, 0.035274f, 0.036503f, 0.037762f, 0.039050f, 0.040370f, 0.041721f, 0.043103f,
0.044518f, 0.045964f, 0.047444f, 0.048956f, 0.050503f, 0.052083f, 0.053699f, 0.055349f,
0.057034f, 0.058755f, 0.060513f, 0.062307f, 0.064139f, 0.066008f, 0.067915f, 0.069861f,
0.071845f, 0.073869f, 0.075933f, 0.078037f, 0.080182f, 0.082369f, 0.084597f, 0.086867f,
0.089180f, 0.091535f, 0.093935f, 0.096378f, 0.098866f, 0.101398f, 0.103977f, 0.106601f,
0.109271f, 0.111988f, 0.114753f, 0.117565f, 0.120426f, 0.123335f, 0.126293f, 0.129301f,
0.132360f, 0.135469f, 0.138629f, 0.141841f, 0.145105f, 0.148421f, 0.151791f, 0.155214f,
0.158691f, 0.162224f, 0.165810f, 0.169453f, 0.173152f, 0.176907f, 0.180720f, 0.184589f,
0.188517f, 0.192504f, 0.196549f, 0.200655f, 0.204820f, 0.209046f, 0.213334f, 0.217682f,
0.222093f, 0.226567f, 0.231104f, 0.235704f, 0.240369f, 0.245099f, 0.249894f, 0.254754f,
0.259681f, 0.264674f, 0.269736f, 0.274864f, 0.280062f, 0.285328f, 0.290664f, 0.296070f,
0.301546f, 0.307094f, 0.312713f, 0.318404f, 0.324168f, 0.330006f, 0.335916f, 0.341902f,
0.347962f, 0.354097f, 0.360309f, 0.366597f, 0.372961f, 0.379403f, 0.385924f, 0.392524f,
0.399202f, 0.405960f, 0.412798f, 0.419718f, 0.426719f, 0.433802f, 0.440967f, 0.448216f,
0.455548f, 0.462965f, 0.470465f, 0.478052f, 0.485725f, 0.493484f, 0.501329f, 0.509263f,
0.517285f, 0.525396f, 0.533595f, 0.541885f, 0.550265f, 0.558736f, 0.567299f, 0.575954f,
0.584702f, 0.593542f, 0.602477f, 0.611507f, 0.620632f, 0.629852f, 0.639168f, 0.648581f,
0.658092f, 0.667700f, 0.677408f, 0.687214f, 0.697120f, 0.707127f, 0.717234f, 0.727443f,
0.737753f, 0.748167f, 0.758685f, 0.769305f, 0.780031f, 0.790861f, 0.801798f, 0.812839f,
0.823989f, 0.835246f, 0.846611f, 0.858085f, 0.869668f, 0.881360f, 0.893164f, 0.905078f,
0.917104f, 0.929242f, 0.941493f, 0.953859f, 0.966338f, 1.000000f, 1.000000f, 1.000000f,
};
} // Anonymous namespace
std::array<float, 4> TSCEntry::BorderColor() const noexcept {
// TODO: Handle SRGB correctly. Using this breaks shadows in some games (Xenoblade).
// if (!srgb_conversion) {
// return border_color;
//}
// return {SRGB_CONVERSION_LUT[srgb_border_color_r], SRGB_CONVERSION_LUT[srgb_border_color_g],
// SRGB_CONVERSION_LUT[srgb_border_color_b], border_color[3]};
return border_color;
}
std::array<float, 4> TSCEntry::SrgbBorderColor() const noexcept {
return {SrgbToLinear(srgb_border_color_r), SrgbToLinear(srgb_border_color_g),
SrgbToLinear(srgb_border_color_b), border_color[3]};
}
float TSCEntry::MaxAnisotropy() const noexcept {
const bool is_suitable_mipmap_filter = mipmap_filter != TextureMipmapFilter::None;
const bool has_regular_lods = min_lod_clamp == 0 && max_lod_clamp >= 256;
@@ -53,16 +80,16 @@ float TSCEntry::MaxAnisotropy() const noexcept {
case Settings::AnisotropyMode::X4:
case Settings::AnisotropyMode::X8:
case Settings::AnisotropyMode::X16:
added_anisotropic = s32(anisotropic_settings) - 1;
case Settings::AnisotropyMode::X32:
case Settings::AnisotropyMode::X64:
added_anisotropic = u32(anisotropic_settings) - 1U;
break;
case Settings::AnisotropyMode::Automatic: {
const u32 resolution_scale = Settings::values.resolution_info.up_scale >>
Settings::values.resolution_info.down_shift;
if (resolution_scale > 1U) {
added_anisotropic = static_cast<s32>(resolution_scale - 1U);
}
case Settings::AnisotropyMode::Automatic:
added_anisotropic = Settings::values.resolution_info.up_scale >> Settings::values.resolution_info.down_shift;
added_anisotropic = (std::max)(added_anisotropic - 1U, 0U);
break;
}
case Settings::AnisotropyMode::None:
return 1.0f; //No use of anisotropy
}
return float(1U << (max_anisotropy + added_anisotropic));
}
-5
View File
@@ -1,6 +1,3 @@
// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
// SPDX-License-Identifier: GPL-3.0-or-later
// SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
@@ -381,8 +378,6 @@ struct TSCEntry {
std::array<float, 4> BorderColor() const noexcept;
std::array<float, 4> SrgbBorderColor() const noexcept;
float MaxAnisotropy() const noexcept;
float MinLod() const {
+130 -71
View File
@@ -7,6 +7,8 @@
#include <algorithm>
#include <bitset>
#include <chrono>
#include <filesystem>
#include <fstream>
#include <optional>
#include <thread>
#include <ankerl/unordered_dense.h>
@@ -16,7 +18,8 @@
#include <fmt/format.h>
#include "common/assert.h"
#include "common/host_memory.h"
#include "common/fs/fs.h"
#include "common/fs/path_util.h"
#include "common/literals.h"
#include <ranges>
#include "common/settings.h"
@@ -394,6 +397,17 @@ std::vector<const char*> ExtensionListForVulkan(
return output;
}
constexpr std::array<char, 8> STATIC_CACHE_MAGIC_NUMBER{'e', 'd', 'e', 'n', 's', 't', 'p', 'c'};
constexpr u32 STATIC_CACHE_VERSION = 1;
std::filesystem::path StaticPipelineCacheFilename() {
const auto shader_dir = Common::FS::GetEdenPath(Common::FS::EdenPath::ShaderDir);
if (!Common::FS::CreateDir(shader_dir)) {
return {};
}
return shader_dir / "vulkan_static_pipelines.bin";
}
} // Anonymous namespace
void Device::RemoveExtension(bool& extension, const std::string& extension_name) {
@@ -509,9 +523,17 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
LOG_WARNING(Render_Vulkan, "Qualcomm drivers require scaled vertex format emulation.");
has_broken_descriptor_aliasing = true;
LOG_WARNING(Render_Vulkan, "Qualcomm drivers have broken descriptor aliasing.");
LOG_WARNING(Render_Vulkan, "Qualcomm drivers have broken custom border color.");
RemoveExtensionFeature(extensions.custom_border_color, features.custom_border_color,
VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME);
LOG_WARNING(Render_Vulkan, "Qualcomm drivers have broken border color swizzle.");
RemoveExtensionFeature(extensions.border_color_swizzle, features.border_color_swizzle,
VK_EXT_BORDER_COLOR_SWIZZLE_EXTENSION_NAME);
LOG_WARNING(Render_Vulkan, "Qualcomm drivers have broken color write enable.");
RemoveExtensionFeature(extensions.color_write_enable, features.color_write_enable,
VK_EXT_COLOR_WRITE_ENABLE_EXTENSION_NAME);
LOG_WARNING(Render_Vulkan, "Qualcomm drivers have broken shader float controls.");
RemoveExtension(extensions.shader_float_controls, VK_KHR_SHADER_FLOAT_CONTROLS_EXTENSION_NAME);
LOG_WARNING(Render_Vulkan, "Qualcomm drivers have broken shader atomic int64.");
RemoveExtensionFeature(extensions.shader_atomic_int64, features.shader_atomic_int64,
VK_KHR_SHADER_ATOMIC_INT64_EXTENSION_NAME);
@@ -612,6 +634,21 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
}
}
if (is_qualcomm) {
const size_t sampler_limit = properties.properties.limits.maxSamplerAllocationCount;
if (sampler_limit > 0) {
constexpr size_t MIN_SAMPLER_BUDGET = 1024U;
const size_t reserved = sampler_limit / 4U;
const size_t derived_budget =
(std::max)(MIN_SAMPLER_BUDGET, sampler_limit - reserved);
sampler_heap_budget = derived_budget;
LOG_WARNING(Render_Vulkan,
"Qualcomm driver reports max {} samplers; reserving {} (25%) and "
"allowing Eden to use {} (75%) to avoid heap exhaustion",
sampler_limit, reserved, sampler_heap_budget);
}
}
if (extensions.sampler_filter_minmax && is_amd) {
// Disable ext_sampler_filter_minmax on AMD GCN4 and lower as it is broken.
if (!features.shader_float16_int8.shaderFloat16) {
@@ -751,15 +788,100 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
vk::Check(vmaCreateAllocator(&allocator_info, &allocator));
owns_static_pipeline_cache = surface != VkSurfaceKHR{};
LoadStaticPipelineCache();
// Initialize GPU logging if enabled
InitializeGPULogging();
}
Device::~Device() {
SaveStaticPipelineCache();
ShutdownGPULogging();
vmaDestroyAllocator(allocator);
}
void Device::LoadStaticPipelineCache() {
const auto create = [this](size_t size, const void* data) {
static_pipeline_cache = logical.CreatePipelineCache({
.sType = VK_STRUCTURE_TYPE_PIPELINE_CACHE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.initialDataSize = size,
.pInitialData = data,
});
};
if (!owns_static_pipeline_cache) {
create(0, nullptr);
return;
}
const auto filename = StaticPipelineCacheFilename();
if (filename.empty()) {
create(0, nullptr);
return;
}
std::vector<char> data;
try {
std::ifstream file(filename, std::ios::binary | std::ios::ate);
if (!file.is_open()) {
create(0, nullptr);
return;
}
file.exceptions(std::ifstream::failbit | std::ifstream::badbit);
const size_t total = static_cast<size_t>(file.tellg());
file.seekg(0, std::ios::beg);
std::array<char, 8> magic{};
u32 version{};
if (total < magic.size() + sizeof(version)) {
create(0, nullptr);
return;
}
file.read(magic.data(), magic.size())
.read(reinterpret_cast<char*>(&version), sizeof(version));
if (magic != STATIC_CACHE_MAGIC_NUMBER || version != STATIC_CACHE_VERSION) {
create(0, nullptr);
return;
}
data.resize(total - magic.size() - sizeof(version));
file.read(data.data(), static_cast<std::streamsize>(data.size()));
} catch (const std::ios_base::failure& e) {
create(0, nullptr);
return;
}
create(data.size(), data.empty() ? nullptr : data.data());
}
void Device::SaveStaticPipelineCache() const {
if (!owns_static_pipeline_cache || !static_pipeline_cache) {
return;
}
const auto filename = StaticPipelineCacheFilename();
if (filename.empty()) {
return;
}
size_t size = 0;
std::vector<char> data;
static_pipeline_cache.Read(&size, nullptr);
if (size == 0) {
return;
}
data.resize(size);
static_pipeline_cache.Read(&size, data.data());
try {
std::ofstream file(filename, std::ios::binary | std::ios::trunc);
file.exceptions(std::ofstream::failbit);
if (!file.is_open()) {
return;
}
file.write(STATIC_CACHE_MAGIC_NUMBER.data(), STATIC_CACHE_MAGIC_NUMBER.size())
.write(reinterpret_cast<const char*>(&STATIC_CACHE_VERSION),
sizeof(STATIC_CACHE_VERSION))
.write(data.data(), static_cast<std::streamsize>(size));
} catch (const std::ios_base::failure& e) {
Common::FS::RemoveFile(filename);
}
}
VkFormat Device::GetSupportedFormat(VkFormat wanted_format, VkFormatFeatureFlags wanted_usage,
FormatType format_type) const {
if (IsFormatSupported(wanted_format, wanted_usage, format_type)) {
@@ -945,7 +1067,6 @@ bool Device::GetSuitability(bool requires_swapchain) {
FOR_EACH_VK_FEATURE_EXT(FEATURE_EXTENSION);
FOR_EACH_VK_EXTENSION(EXTENSION);
FOR_EACH_VK_PLATFORM_EXTENSION(EXTENSION);
extensions.depth_stencil_resolve =
extensions.depth_stencil_resolve &&
@@ -964,13 +1085,6 @@ bool Device::GetSuitability(bool requires_swapchain) {
extensions.robustness_2 = false;
}
#ifdef __ANDROID__
if (extensions.external_memory_ahb && !extensions.queue_family_foreign) {
loaded_extensions.erase(VK_ANDROID_EXTERNAL_MEMORY_ANDROID_HARDWARE_BUFFER_EXTENSION_NAME);
extensions.external_memory_ahb = false;
}
#endif
#undef FEATURE_EXTENSION
#undef EXTENSION
@@ -1132,26 +1246,6 @@ bool Device::GetSuitability(bool requires_swapchain) {
VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MAINTENANCE_5_PROPERTIES_KHR;
SetNext(next, properties.maintenance5);
}
if (extensions.custom_border_color) {
properties.custom_border_color.sType =
VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CUSTOM_BORDER_COLOR_PROPERTIES_EXT;
SetNext(next, properties.custom_border_color);
}
if (extensions.external_memory_host) {
properties.external_memory_host.sType =
VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_EXTERNAL_MEMORY_HOST_PROPERTIES_EXT;
SetNext(next, properties.external_memory_host);
}
if (extensions.maintenance3 || instance_version >= VK_API_VERSION_1_1) {
properties.maintenance3.sType =
VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MAINTENANCE_3_PROPERTIES;
SetNext(next, properties.maintenance3);
}
if (extensions.maintenance4 || features.maintenance4.maintenance4) {
properties.maintenance4.sType =
VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MAINTENANCE_4_PROPERTIES;
SetNext(next, properties.maintenance4);
}
// Perform the property fetch.
physical.GetProperties2(properties2);
@@ -1248,7 +1342,9 @@ void Device::RemoveUnsuitableExtensions() {
// VK_EXT_border_color_swizzle
if (extensions.border_color_swizzle) {
extensions.border_color_swizzle =
extensions.custom_border_color && features.border_color_swizzle.borderColorSwizzle;
extensions.custom_border_color &&
features.border_color_swizzle.borderColorSwizzle &&
features.border_color_swizzle.borderColorSwizzleFromImage;
}
RemoveExtensionFeatureIfUnsuitable(extensions.border_color_swizzle,
features.border_color_swizzle,
@@ -1494,26 +1590,11 @@ void Device::SetupFamilies(VkSurfaceKHR surface) {
}
}
bool Device::TryReserveCustomBorderColorSamplers(size_t count) const {
const size_t limit = properties.custom_border_color.maxCustomBorderColorSamplers;
if (limit == 0) {
return true;
std::optional<size_t> Device::GetSamplerHeapBudget() const {
if (sampler_heap_budget == 0) {
return std::nullopt;
}
size_t used = custom_border_color_samplers_used.load(std::memory_order_relaxed);
while (used + count <= limit) {
if (custom_border_color_samplers_used.compare_exchange_weak(
used, used + count, std::memory_order_relaxed, std::memory_order_relaxed)) {
return true;
}
}
return false;
}
void Device::ReleaseCustomBorderColorSamplers(size_t count) const {
if (count == 0) {
return;
}
custom_border_color_samplers_used.fetch_sub(count, std::memory_order_relaxed);
return sampler_heap_budget;
}
u64 Device::GetDeviceMemoryUsage() const {
@@ -1539,27 +1620,12 @@ void Device::CollectPhysicalMemoryInfo() {
device_access_memory = 0;
u64 device_initial_usage = 0;
u64 local_memory = 0;
const auto heap_has_usable_type = [&mem_properties](size_t heap) {
for (u32 index = 0; index < mem_properties.memoryTypeCount; ++index) {
if (mem_properties.memoryTypes[index].heapIndex != heap) {
continue;
}
if ((mem_properties.memoryTypes[index].propertyFlags &
VK_MEMORY_PROPERTY_PROTECTED_BIT) == 0) {
return true;
}
}
return false;
};
for (size_t element = 0; element < num_properties; ++element) {
const bool is_heap_local =
(mem_properties.memoryHeaps[element].flags & VK_MEMORY_HEAP_DEVICE_LOCAL_BIT) != 0;
if (!is_integrated && !is_heap_local) {
continue;
}
if (!heap_has_usable_type(element)) {
continue;
}
valid_heap_memory.push_back(element);
if (is_heap_local) {
local_memory += mem_properties.memoryHeaps[element].size;
@@ -1571,13 +1637,6 @@ void Device::CollectPhysicalMemoryInfo() {
}
device_access_memory += mem_properties.memoryHeaps[element].size;
}
const u64 committed_backing = Common::GetCommittedBackingSize();
if (committed_backing != 0) {
LOG_INFO(Render_Vulkan, "Discounting {} MiB of guest memory committed by the host",
committed_backing >> 20);
local_memory -= std::min(local_memory, committed_backing);
device_access_memory -= std::min(device_access_memory, committed_backing);
}
if (is_integrated) {
const s64 available_memory = static_cast<s64>(device_access_memory - device_initial_usage);
const u64 memory_size = Settings::values.vram_usage_mode.GetValue() == Settings::VramUsageMode::Aggressive ? 6_GiB : 4_GiB;
+23 -55
View File
@@ -6,7 +6,6 @@
#pragma once
#include <atomic>
#include <optional>
#include <set>
#include <span>
@@ -85,7 +84,6 @@ VK_DEFINE_HANDLE(VmaAllocator)
EXTENSION(EXT, CONDITIONAL_RENDERING, conditional_rendering) \
EXTENSION(EXT, CONSERVATIVE_RASTERIZATION, conservative_rasterization) \
EXTENSION(EXT, DEPTH_RANGE_UNRESTRICTED, depth_range_unrestricted) \
EXTENSION(EXT, EXTERNAL_MEMORY_HOST, external_memory_host) \
EXTENSION(EXT, MEMORY_BUDGET, memory_budget) \
EXTENSION(EXT, ROBUSTNESS_2, robustness_2) \
EXTENSION(EXT, SAMPLER_FILTER_MINMAX, sampler_filter_minmax) \
@@ -117,14 +115,6 @@ VK_DEFINE_HANDLE(VmaAllocator)
EXTENSION(IMG, FILTER_CUBIC, filter_cubic_img) \
EXTENSION(QCOM, FILTER_CUBIC_WEIGHTS, filter_cubic_weights)
#ifdef __ANDROID__
#define FOR_EACH_VK_PLATFORM_EXTENSION(EXTENSION) \
EXTENSION(EXT, QUEUE_FAMILY_FOREIGN, queue_family_foreign) \
EXTENSION(ANDROID, EXTERNAL_MEMORY_ANDROID_HARDWARE_BUFFER, external_memory_ahb)
#else
#define FOR_EACH_VK_PLATFORM_EXTENSION(EXTENSION)
#endif
// Define extensions which must be supported.
#define FOR_EACH_VK_MANDATORY_EXTENSION(EXTENSION_NAME) \
EXTENSION_NAME(VK_EXT_VERTEX_ATTRIBUTE_DIVISOR_EXTENSION_NAME) \
@@ -281,6 +271,10 @@ public:
return physical;
}
VkPipelineCache StaticPipelineCache() const noexcept {
return *static_pipeline_cache;
}
/// Returns the main graphics queue.
vk::Queue GetGraphicsQueue() const {
return graphics_queue;
@@ -709,18 +703,20 @@ FN_MAX_LIMIT_LIST
return features.transform_feedback.geometryStreams;
}
/// Returns true if custom border colors can be created without a format.
bool IsCustomBorderColorUsable() const {
return extensions.custom_border_color &&
features.custom_border_color.customBorderColors &&
features.custom_border_color.customBorderColorWithoutFormat;
/// Returns true if the device supports VK_EXT_custom_border_color.
bool IsExtCustomBorderColorSupported() const {
return extensions.custom_border_color;
}
/// Takes budget for samplers carrying a custom border color, false when exhausted.
bool TryReserveCustomBorderColorSamplers(size_t count) const;
/// Returns true if customBorderColors feature is available.
bool IsCustomBorderColorsSupported() const {
return features.custom_border_color.customBorderColors;
}
/// Gives back budget taken by TryReserveCustomBorderColorSamplers.
void ReleaseCustomBorderColorSamplers(size_t count) const;
/// Returns true if customBorderColorWithoutFormat feature is available.
bool IsCustomBorderColorWithoutFormatSupported() const {
return features.custom_border_color.customBorderColorWithoutFormat;
}
/// Returns true if the device supports VK_EXT_color_write_enable.
bool IsExtColorWriteEnableSupported() const {
@@ -732,12 +728,6 @@ FN_MAX_LIMIT_LIST
return extensions.border_color_swizzle;
}
/// Returns true if samplers must be carried with border color swizzle mapping.
bool NeedsBorderColorSwizzleMapping() const {
return extensions.border_color_swizzle &&
!features.border_color_swizzle.borderColorSwizzleFromImage;
}
/// Returns true if borderColorSwizzleFromImage is available.
bool IsBorderColorSwizzleFromImageSupported() const {
return features.border_color_swizzle.borderColorSwizzleFromImage;
@@ -886,30 +876,6 @@ FN_MAX_LIMIT_LIST
return extensions.conditional_rendering;
}
bool IsExtExternalMemoryHostSupported() const {
return extensions.external_memory_host;
}
bool IsExtExternalMemoryAhbSupported() const {
#ifdef __ANDROID__
return extensions.external_memory_ahb && extensions.queue_family_foreign;
#else
return false;
#endif
}
u64 GetMinImportedHostPointerAlignment() const {
return properties.external_memory_host.minImportedHostPointerAlignment;
}
u64 GetMaxBufferSize() const {
return properties.maintenance4.maxBufferSize;
}
u64 GetMaxMemoryAllocationSize() const {
return properties.maintenance3.maxMemoryAllocationSize;
}
bool IsExtAstcDecodeModeSupported() const {
return extensions.astc_decode_mode;
}
@@ -957,6 +923,8 @@ FN_MAX_LIMIT_LIST
return has_broken_parallel_compiling;
}
std::optional<size_t> GetSamplerHeapBudget() const;
/// Returns the vendor name reported from Vulkan.
std::string_view GetVendorName() const {
return properties.driver.driverName;
@@ -1154,6 +1122,9 @@ private:
/// Returns true if the device natively supports blitting depth stencil images.
bool TestDepthStencilBlits(VkFormat format) const;
void LoadStaticPipelineCache();
void SaveStaticPipelineCache() const;
private:
VkInstance instance; ///< Vulkan instance.
VmaAllocator allocator; ///< VMA allocator.
@@ -1162,6 +1133,8 @@ private:
vk::Device logical; ///< Logical device.
vk::Queue graphics_queue; ///< Main graphics queue.
vk::Queue present_queue; ///< Main present queue.
vk::PipelineCache static_pipeline_cache;
bool owns_static_pipeline_cache{};
u32 instance_version{}; ///< Vulkan instance version.
u32 graphics_family{}; ///< Main graphics queue family index.
u32 present_family{}; ///< Main present queue family index.
@@ -1176,7 +1149,6 @@ private:
FOR_EACH_VK_FEATURE_1_4(FEATURE);
FOR_EACH_VK_FEATURE_EXT(FEATURE);
FOR_EACH_VK_EXTENSION(EXTENSION);
FOR_EACH_VK_PLATFORM_EXTENSION(EXTENSION);
#undef EXTENSION
#undef FEATURE
@@ -1208,12 +1180,8 @@ private:
VkPhysicalDeviceDescriptorBufferPropertiesEXT descriptor_buffer{};
VkPhysicalDeviceSubgroupSizeControlProperties subgroup_size_control{};
VkPhysicalDeviceTransformFeedbackPropertiesEXT transform_feedback{};
VkPhysicalDeviceMaintenance3Properties maintenance3{};
VkPhysicalDeviceMaintenance4Properties maintenance4{};
VkPhysicalDeviceMaintenance5PropertiesKHR maintenance5{};
VkPhysicalDeviceDepthStencilResolveProperties depth_stencil_resolve{};
VkPhysicalDeviceCustomBorderColorPropertiesEXT custom_border_color{};
VkPhysicalDeviceExternalMemoryHostPropertiesEXT external_memory_host{};
VkPhysicalDeviceProperties properties{};
};
@@ -1252,7 +1220,7 @@ private:
bool dynamic_state3_alpha_to_coverage{};
bool dynamic_state3_alpha_to_one{};
bool supports_conditional_barriers{}; ///< Allows barriers in conditional control flow.
mutable std::atomic<size_t> custom_border_color_samplers_used{};
size_t sampler_heap_budget{}; ///< Sampler budget for buggy drivers (0 = unlimited).
u64 device_access_memory{}; ///< Total size of device local memory in bytes.
u32 sets_per_pool{}; ///< Sets per Description Pool
NvidiaArchitecture nvidia_arch{NvidiaArchitecture::Arch_AmpereOrNewer};
@@ -25,34 +25,9 @@
#include "video_core/gpu_logging/gpu_logging.h"
#include "common/settings.h"
#ifdef __ANDROID__
#include <android/hardware_buffer.h>
#endif
namespace Vulkan {
namespace {
[[nodiscard]] std::optional<u32> FindImportMemoryType(
const VkPhysicalDeviceMemoryProperties &props, u32 type_mask) {
const auto find = [&](VkMemoryPropertyFlags wanted) -> std::optional<u32> {
for (u32 i = 0; i < props.memoryTypeCount; ++i) {
if (((type_mask >> i) & 1u) != 0 &&
(props.memoryTypes[i].propertyFlags & wanted) == wanted) {
return i;
}
}
return std::nullopt;
};
auto type_index = find(VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
VK_MEMORY_PROPERTY_HOST_COHERENT_BIT |
VK_MEMORY_PROPERTY_HOST_CACHED_BIT);
if (!type_index) {
type_index = find(VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
VK_MEMORY_PROPERTY_HOST_COHERENT_BIT);
}
return type_index;
}
// Helpers translating MemoryUsage to flags/usage
[[maybe_unused]] VkMemoryPropertyFlags MemoryUsagePropertyFlags(MemoryUsage usage) {
@@ -225,300 +200,6 @@ namespace Vulkan {
size = 0;
}
HostMemoryImport::HostMemoryImport(const Device &device_, void *base, size_t size,
std::span<AHardwareBuffer *const> hardware_buffers,
size_t hardware_buffer_window, size_t hardware_buffer_base)
: device{device_} {
if (ImportHardwareBuffers(hardware_buffers, hardware_buffer_window, hardware_buffer_base,
size)) {
return;
}
if (device.IsTiler()) {
return;
}
if (!hardware_buffers.empty()) {
return;
}
if (ImportHostPointer(base, size)) {
return;
}
LOG_INFO(Render_Vulkan, "Unified memory disabled, no host memory import path");
}
bool HostMemoryImport::ImportHostPointer(void *base, size_t size) {
if (!device.IsExtExternalMemoryHostSupported()) {
return false;
}
const u64 alignment = device.GetMinImportedHostPointerAlignment();
if (alignment == 0 || !Common::IsAligned(reinterpret_cast<uintptr_t>(base), alignment) ||
!Common::IsAligned(size, alignment)) {
return false;
}
using namespace Common::Literals;
constexpr VkDeviceSize DesktopWindowSize = 4_GiB;
VkDeviceSize candidate_window = DesktopWindowSize;
const u64 max_buffer_size = device.GetMaxBufferSize();
if (max_buffer_size != 0 && max_buffer_size < candidate_window) {
candidate_window = max_buffer_size;
}
const u64 max_allocation_size = device.GetMaxMemoryAllocationSize();
if (max_allocation_size != 0 && max_allocation_size < candidate_window) {
candidate_window = max_allocation_size;
}
candidate_window = Common::AlignDown(candidate_window, alignment);
if (candidate_window == 0) {
return false;
}
window_size = candidate_window;
const auto &logical = device.GetLogical();
const auto memory_props = device.GetPhysical().GetMemoryProperties().memoryProperties;
for (size_t offset = 0; offset < size; offset += window_size) {
u8 *const window_base = static_cast<u8 *>(base) + offset;
const VkDeviceSize window_len =
(std::min)(static_cast<VkDeviceSize>(size - offset), window_size);
VkMemoryHostPointerPropertiesEXT host_props{
.sType = VK_STRUCTURE_TYPE_MEMORY_HOST_POINTER_PROPERTIES_EXT,
.pNext = nullptr,
.memoryTypeBits = 0,
};
if (logical.GetMemoryHostPointerPropertiesEXT(
VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_ALLOCATION_BIT_EXT, window_base,
&host_props) != VK_SUCCESS ||
host_props.memoryTypeBits == 0) {
break;
}
const VkExternalMemoryBufferCreateInfo external_info{
.sType = VK_STRUCTURE_TYPE_EXTERNAL_MEMORY_BUFFER_CREATE_INFO,
.pNext = nullptr,
.handleTypes = VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_ALLOCATION_BIT_EXT,
};
const VkBufferCreateInfo buffer_ci{
.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
.pNext = &external_info,
.flags = 0,
.size = window_len,
.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT,
.sharingMode = VK_SHARING_MODE_EXCLUSIVE,
.queueFamilyIndexCount = 0,
.pQueueFamilyIndices = nullptr,
};
VkBuffer new_buffer{};
if (logical.CreateBufferRaw(buffer_ci, &new_buffer) != VK_SUCCESS) {
break;
}
const VkMemoryRequirements requirements =
logical.GetBufferMemoryRequirements(new_buffer);
const u32 type_mask = requirements.memoryTypeBits & host_props.memoryTypeBits;
if (type_mask == 0 || requirements.size > window_len) {
logical.DestroyBufferRaw(new_buffer);
break;
}
const auto type_index = FindImportMemoryType(memory_props, type_mask);
if (!type_index) {
logical.DestroyBufferRaw(new_buffer);
break;
}
const u32 heap_index = memory_props.memoryTypes[*type_index].heapIndex;
const VkDeviceSize heap_size = memory_props.memoryHeaps[heap_index].size;
if (imported_size + window_len > heap_size / 2) {
logical.DestroyBufferRaw(new_buffer);
break;
}
const VkImportMemoryHostPointerInfoEXT import_info{
.sType = VK_STRUCTURE_TYPE_IMPORT_MEMORY_HOST_POINTER_INFO_EXT,
.pNext = nullptr,
.handleType = VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_ALLOCATION_BIT_EXT,
.pHostPointer = window_base,
};
const VkMemoryAllocateInfo alloc_info{
.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO,
.pNext = &import_info,
.allocationSize = window_len,
.memoryTypeIndex = *type_index,
};
vk::DeviceMemory memory = logical.TryAllocateMemory(alloc_info);
if (!memory) {
logical.DestroyBufferRaw(new_buffer);
break;
}
if (logical.BindBufferMemory(new_buffer, *memory, 0) != VK_SUCCESS) {
logical.DestroyBufferRaw(new_buffer);
break;
}
windows.push_back(Window{
.memory = std::move(memory),
.buffer = new_buffer,
});
imported_size += static_cast<size_t>(window_len);
}
if (windows.empty()) {
return false;
}
return true;
}
bool HostMemoryImport::ImportHardwareBuffers(
[[maybe_unused]] std::span<AHardwareBuffer *const> hardware_buffers,
[[maybe_unused]] size_t hardware_buffer_window,
[[maybe_unused]] size_t hardware_buffer_base, [[maybe_unused]] size_t size) {
#ifdef __ANDROID__
if (hardware_buffers.empty() || hardware_buffer_window == 0 ||
!device.IsExtExternalMemoryAhbSupported()) {
return false;
}
const u64 max_allocation_size = device.GetMaxMemoryAllocationSize();
if (max_allocation_size != 0 && hardware_buffer_window > max_allocation_size) {
return false;
}
if (hardware_buffer_base >= size) {
return false;
}
const auto &logical = device.GetLogical();
const auto memory_props = device.GetPhysical().GetMemoryProperties().memoryProperties;
window_size = hardware_buffer_window;
base_offset = hardware_buffer_base;
const auto import_all = [&](VkBufferUsageFlags usage, bool want_address) {
for (size_t i = 0; i < hardware_buffers.size(); ++i) {
const size_t offset = hardware_buffer_base + i * hardware_buffer_window;
if (offset >= size) {
break;
}
const VkDeviceSize window_len = (std::min)(
static_cast<VkDeviceSize>(size - offset),
static_cast<VkDeviceSize>(hardware_buffer_window));
VkAndroidHardwareBufferPropertiesANDROID ahb_props{
.sType = VK_STRUCTURE_TYPE_ANDROID_HARDWARE_BUFFER_PROPERTIES_ANDROID,
.pNext = nullptr,
.allocationSize = 0,
.memoryTypeBits = 0,
};
if (logical.GetAndroidHardwareBufferPropertiesANDROID(hardware_buffers[i],
&ahb_props) != VK_SUCCESS ||
ahb_props.memoryTypeBits == 0 || ahb_props.allocationSize < window_len) {
break;
}
const VkExternalMemoryBufferCreateInfo external_info{
.sType = VK_STRUCTURE_TYPE_EXTERNAL_MEMORY_BUFFER_CREATE_INFO,
.pNext = nullptr,
.handleTypes =
VK_EXTERNAL_MEMORY_HANDLE_TYPE_ANDROID_HARDWARE_BUFFER_BIT_ANDROID,
};
const VkBufferCreateInfo buffer_ci{
.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
.pNext = &external_info,
.flags = 0,
.size = window_len,
.usage = usage,
.sharingMode = VK_SHARING_MODE_EXCLUSIVE,
.queueFamilyIndexCount = 0,
.pQueueFamilyIndices = nullptr,
};
VkBuffer new_buffer{};
if (logical.CreateBufferRaw(buffer_ci, &new_buffer) != VK_SUCCESS) {
break;
}
const VkMemoryRequirements requirements =
logical.GetBufferMemoryRequirements(new_buffer);
const u32 type_mask = requirements.memoryTypeBits & ahb_props.memoryTypeBits;
if (type_mask == 0 || requirements.size > ahb_props.allocationSize) {
logical.DestroyBufferRaw(new_buffer);
break;
}
const auto type_index = FindImportMemoryType(memory_props, type_mask);
if (!type_index) {
logical.DestroyBufferRaw(new_buffer);
break;
}
const VkImportAndroidHardwareBufferInfoANDROID import_info{
.sType = VK_STRUCTURE_TYPE_IMPORT_ANDROID_HARDWARE_BUFFER_INFO_ANDROID,
.pNext = nullptr,
.buffer = hardware_buffers[i],
};
const VkMemoryDedicatedAllocateInfo dedicated_info{
.sType = VK_STRUCTURE_TYPE_MEMORY_DEDICATED_ALLOCATE_INFO,
.pNext = &import_info,
.image = VK_NULL_HANDLE,
.buffer = new_buffer,
};
const VkMemoryAllocateFlagsInfo flags_info{
.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_FLAGS_INFO,
.pNext = &dedicated_info,
.flags = VK_MEMORY_ALLOCATE_DEVICE_ADDRESS_BIT,
.deviceMask = 0,
};
const void *alloc_next = &dedicated_info;
if (want_address) {
alloc_next = &flags_info;
}
const VkMemoryAllocateInfo alloc_info{
.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO,
.pNext = alloc_next,
.allocationSize = ahb_props.allocationSize,
.memoryTypeIndex = *type_index,
};
vk::DeviceMemory memory = logical.TryAllocateMemory(alloc_info);
if (!memory) {
logical.DestroyBufferRaw(new_buffer);
break;
}
if (logical.BindBufferMemory(new_buffer, *memory, 0) != VK_SUCCESS) {
logical.DestroyBufferRaw(new_buffer);
break;
}
VkDeviceAddress address = 0;
if (want_address) {
address = logical.GetBufferDeviceAddress(new_buffer);
}
windows.push_back(Window{
.memory = std::move(memory),
.buffer = new_buffer,
.address = address,
});
imported_size += static_cast<size_t>(window_len);
}
return !windows.empty();
};
constexpr VkBufferUsageFlags TransferUsage =
VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT;
VkBufferUsageFlags shader_usage = TransferUsage |
VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT |
VK_BUFFER_USAGE_STORAGE_TEXEL_BUFFER_BIT | VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT |
VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | VK_BUFFER_USAGE_INDEX_BUFFER_BIT |
VK_BUFFER_USAGE_VERTEX_BUFFER_BIT | VK_BUFFER_USAGE_INDIRECT_BUFFER_BIT;
const bool want_address = device.IsBufferDeviceAddressSupported();
if (want_address) {
shader_usage |= VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT;
}
bindable = import_all(shader_usage, want_address);
if (!bindable) {
imported_size = 0;
import_all(TransferUsage, false);
}
if (windows.empty()) {
window_size = 0;
base_offset = 0;
return false;
}
foreign_ownership = true;
return true;
#else
return false;
#endif
}
HostMemoryImport::~HostMemoryImport() {
for (Window &window : windows) {
if (window.buffer != VK_NULL_HANDLE) {
device.GetLogical().DestroyBufferRaw(window.buffer);
}
}
}
MemoryAllocator::MemoryAllocator(const Device &device_)
: device{device_}, allocator{device.GetAllocator()},
properties{device_.GetPhysical().GetMemoryProperties().memoryProperties},
@@ -15,8 +15,6 @@
#include "video_core/vulkan_common/vulkan_wrapper.h"
#include "video_core/vulkan_common/vma.h"
struct AHardwareBuffer;
namespace Vulkan {
class Device;
@@ -86,76 +84,6 @@ namespace Vulkan {
void *mapped_ptr{}; ///< Optional persistent mapped pointer
};
class HostMemoryImport {
public:
explicit HostMemoryImport(const Device &device_, void *base, size_t size,
std::span<AHardwareBuffer *const> hardware_buffers,
size_t hardware_buffer_window, size_t hardware_buffer_base);
~HostMemoryImport();
HostMemoryImport(const HostMemoryImport &) = delete;
HostMemoryImport &operator=(const HostMemoryImport &) = delete;
[[nodiscard]] bool IsValid() const noexcept {
return !windows.empty();
}
[[nodiscard]] size_t GetSize() const noexcept {
return imported_size;
}
[[nodiscard]] size_t GetBaseOffset() const noexcept {
return base_offset;
}
[[nodiscard]] bool NeedsForeignOwnershipTransfer() const noexcept {
return foreign_ownership;
}
[[nodiscard]] VkDeviceSize GetWindowSize() const noexcept {
return window_size;
}
[[nodiscard]] VkBuffer GetWindowBuffer(size_t index) const noexcept {
return windows[index].buffer;
}
[[nodiscard]] VkDeviceAddress GetWindowAddress(size_t index) const noexcept {
return windows[index].address;
}
[[nodiscard]] size_t GetWindowCount() const noexcept {
return windows.size();
}
[[nodiscard]] bool IsBindable() const noexcept {
return bindable;
}
private:
struct Window {
vk::DeviceMemory memory;
VkBuffer buffer{};
VkDeviceAddress address{};
};
bool ImportHostPointer(void *base, size_t size);
bool ImportHardwareBuffers(std::span<AHardwareBuffer *const> hardware_buffers,
size_t hardware_buffer_window, size_t hardware_buffer_base,
size_t size);
const Device &device;
std::vector<Window> windows;
VkDeviceSize window_size{};
size_t imported_size{};
size_t base_offset{};
bool foreign_ownership{};
bool bindable{};
};
/// Memory allocator container.
/// Allocates and releases memory allocations on demand.
class MemoryAllocator {
@@ -217,16 +217,12 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept {
X(vkGetBufferMemoryRequirements2);
X(vkGetDeviceQueue);
X(vkGetEventStatus);
X(vkGetMemoryHostPointerPropertiesEXT);
X(vkGetFenceStatus);
X(vkGetImageMemoryRequirements);
X(vkGetPipelineCacheData);
X(vkGetMemoryFdKHR);
#ifdef _WIN32
X(vkGetMemoryWin32HandleKHR);
#endif
#ifdef __ANDROID__
X(vkGetAndroidHardwareBufferPropertiesANDROID);
#endif
X(vkGetQueryPoolResults);
X(vkGetPipelineExecutablePropertiesKHR);
@@ -333,16 +333,12 @@ struct DeviceDispatch : InstanceDispatch {
PFN_vkGetBufferMemoryRequirements2 vkGetBufferMemoryRequirements2{};
PFN_vkGetDeviceQueue vkGetDeviceQueue{};
PFN_vkGetEventStatus vkGetEventStatus{};
PFN_vkGetMemoryHostPointerPropertiesEXT vkGetMemoryHostPointerPropertiesEXT{};
PFN_vkGetFenceStatus vkGetFenceStatus{};
PFN_vkGetImageMemoryRequirements vkGetImageMemoryRequirements{};
PFN_vkGetPipelineCacheData vkGetPipelineCacheData{};
PFN_vkGetMemoryFdKHR vkGetMemoryFdKHR{};
#ifdef _WIN32
PFN_vkGetMemoryWin32HandleKHR vkGetMemoryWin32HandleKHR{};
#endif
#ifdef __ANDROID__
PFN_vkGetAndroidHardwareBufferPropertiesANDROID vkGetAndroidHardwareBufferPropertiesANDROID{};
#endif
PFN_vkGetPipelineExecutablePropertiesKHR vkGetPipelineExecutablePropertiesKHR{};
PFN_vkGetPipelineExecutableStatisticsKHR vkGetPipelineExecutableStatisticsKHR{};
@@ -1089,34 +1085,6 @@ public:
VkMemoryRequirements GetBufferMemoryRequirements(VkBuffer buffer,
void* pnext = nullptr) const noexcept;
VkResult GetMemoryHostPointerPropertiesEXT(
VkExternalMemoryHandleTypeFlagBits handle_type, const void* host_pointer,
VkMemoryHostPointerPropertiesEXT* out_properties) const noexcept {
return dld->vkGetMemoryHostPointerPropertiesEXT(handle, handle_type, host_pointer,
out_properties);
}
#ifdef __ANDROID__
VkResult GetAndroidHardwareBufferPropertiesANDROID(
const struct AHardwareBuffer* buffer,
VkAndroidHardwareBufferPropertiesANDROID* out_properties) const noexcept {
return dld->vkGetAndroidHardwareBufferPropertiesANDROID(handle, buffer, out_properties);
}
#endif
VkResult CreateBufferRaw(const VkBufferCreateInfo& ci, VkBuffer* out_buffer) const noexcept {
return dld->vkCreateBuffer(handle, &ci, nullptr, out_buffer);
}
void DestroyBufferRaw(VkBuffer buffer) const noexcept {
dld->vkDestroyBuffer(handle, buffer, nullptr);
}
VkResult BindBufferMemory(VkBuffer buffer, VkDeviceMemory memory,
VkDeviceSize offset) const noexcept {
return dld->vkBindBufferMemory(handle, buffer, memory, offset);
}
VkMemoryRequirements GetImageMemoryRequirements(VkImage image) const noexcept;
std::vector<VkPipelineExecutablePropertiesKHR> GetPipelineExecutablePropertiesKHR(