early-access version 1265

This commit is contained in:
pineappleEA 2020-12-30 08:25:25 +00:00
parent 1ecd107637
commit d363a1bef6
25 changed files with 1222 additions and 114 deletions

View file

@ -1,7 +1,7 @@
yuzu emulator early access
=============
This is the source code for early-access 1264.
This is the source code for early-access 1265.
## Legal Notice

View file

@ -11,8 +11,9 @@
namespace Service::Nvidia::Devices {
nvhost_nvdec::nvhost_nvdec(Core::System& system, std::shared_ptr<nvmap> nvmap_dev)
: nvhost_nvdec_common(system, std::move(nvmap_dev)) {}
nvhost_nvdec::nvhost_nvdec(Core::System& system, std::shared_ptr<nvmap> nvmap_dev,
SyncpointManager& syncpoint_manager)
: nvhost_nvdec_common(system, std::move(nvmap_dev), syncpoint_manager) {}
nvhost_nvdec::~nvhost_nvdec() = default;
NvResult nvhost_nvdec::Ioctl1(Ioctl command, const std::vector<u8>& input,

View file

@ -11,7 +11,8 @@ namespace Service::Nvidia::Devices {
class nvhost_nvdec final : public nvhost_nvdec_common {
public:
explicit nvhost_nvdec(Core::System& system, std::shared_ptr<nvmap> nvmap_dev);
explicit nvhost_nvdec(Core::System& system, std::shared_ptr<nvmap> nvmap_dev,
SyncpointManager& syncpoint_manager);
~nvhost_nvdec() override;
NvResult Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;

View file

@ -11,6 +11,7 @@
#include "core/core.h"
#include "core/hle/service/nvdrv/devices/nvhost_nvdec_common.h"
#include "core/hle/service/nvdrv/devices/nvmap.h"
#include "core/hle/service/nvdrv/syncpoint_manager.h"
#include "core/memory.h"
#include "video_core/memory_manager.h"
#include "video_core/renderer_base.h"
@ -36,8 +37,9 @@ std::size_t WriteVectors(std::vector<u8>& dst, const std::vector<T>& src, std::s
}
} // Anonymous namespace
nvhost_nvdec_common::nvhost_nvdec_common(Core::System& system, std::shared_ptr<nvmap> nvmap_dev)
: nvdevice(system), nvmap_dev(std::move(nvmap_dev)) {}
nvhost_nvdec_common::nvhost_nvdec_common(Core::System& system, std::shared_ptr<nvmap> nvmap_dev,
SyncpointManager& syncpoint_manager)
: nvdevice(system), nvmap_dev(std::move(nvmap_dev)), syncpoint_manager(syncpoint_manager) {}
nvhost_nvdec_common::~nvhost_nvdec_common() = default;
NvResult nvhost_nvdec_common::SetNVMAPfd(const std::vector<u8>& input) {
@ -71,10 +73,15 @@ NvResult nvhost_nvdec_common::Submit(const std::vector<u8>& input, std::vector<u
offset = SpliceVectors(input, wait_checks, params.syncpoint_count, offset);
offset = SpliceVectors(input, fences, params.fence_count, offset);
// TODO(ameerj): For async gpu, utilize fences for syncpoint 'max' increment
auto& gpu = system.GPU();
if (gpu.UseNvdec()) {
for (std::size_t i = 0; i < syncpt_increments.size(); i++) {
const SyncptIncr& syncpt_incr = syncpt_increments[i];
fences[i].id = syncpt_incr.id;
fences[i].value =
syncpoint_manager.IncreaseSyncpoint(syncpt_incr.id, syncpt_incr.increments);
}
}
for (const auto& cmd_buffer : command_buffers) {
auto object = nvmap_dev->GetObject(cmd_buffer.memory_id);
ASSERT_OR_EXECUTE(object, return NvResult::InvalidState;);
@ -89,7 +96,13 @@ NvResult nvhost_nvdec_common::Submit(const std::vector<u8>& input, std::vector<u
cmdlist.size() * sizeof(u32));
gpu.PushCommandBuffer(cmdlist);
}
if (gpu.UseNvdec()) {
fences[0].value = syncpoint_manager.IncreaseSyncpoint(fences[0].id, 1);
Tegra::ChCommandHeaderList cmdlist{{(4 << 28) | fences[0].id}};
gpu.PushCommandBuffer(cmdlist);
}
std::memcpy(output.data(), &params, sizeof(IoctlSubmit));
// Some games expect command_buffers to be written back
offset = sizeof(IoctlSubmit);
@ -98,6 +111,7 @@ NvResult nvhost_nvdec_common::Submit(const std::vector<u8>& input, std::vector<u
offset = WriteVectors(output, reloc_shifts, offset);
offset = WriteVectors(output, syncpt_increments, offset);
offset = WriteVectors(output, wait_checks, offset);
offset = WriteVectors(output, fences, offset);
return NvResult::Success;
}
@ -107,9 +121,10 @@ NvResult nvhost_nvdec_common::GetSyncpoint(const std::vector<u8>& input, std::ve
std::memcpy(&params, input.data(), sizeof(IoctlGetSyncpoint));
LOG_DEBUG(Service_NVDRV, "called GetSyncpoint, id={}", params.param);
// We found that implementing this causes deadlocks with async gpu, along with degraded
// performance. TODO: RE the nvdec async implementation
params.value = 0;
if (device_syncpoints[params.param] == 0 && system.GPU().UseNvdec()) {
device_syncpoints[params.param] = syncpoint_manager.AllocateSyncpoint();
}
params.value = device_syncpoints[params.param];
std::memcpy(output.data(), &params, sizeof(IoctlGetSyncpoint));
return NvResult::Success;

View file

@ -10,12 +10,16 @@
#include "common/swap.h"
#include "core/hle/service/nvdrv/devices/nvdevice.h"
namespace Service::Nvidia::Devices {
namespace Service::Nvidia {
class SyncpointManager;
namespace Devices {
class nvmap;
class nvhost_nvdec_common : public nvdevice {
public:
explicit nvhost_nvdec_common(Core::System& system, std::shared_ptr<nvmap> nvmap_dev);
explicit nvhost_nvdec_common(Core::System& system, std::shared_ptr<nvmap> nvmap_dev,
SyncpointManager& syncpoint_manager);
~nvhost_nvdec_common() override;
protected:
@ -158,8 +162,10 @@ protected:
u32_le submit_timeout{};
bool vic_device{};
std::shared_ptr<nvmap> nvmap_dev;
SyncpointManager& syncpoint_manager;
std::array<u32, MaxSyncPoints> device_syncpoints{};
// This is expected to be ordered, therefore we must use a map, not unordered_map
std::map<GPUVAddr, BufferMap> buffer_mappings;
};
}; // namespace Service::Nvidia::Devices
}; // namespace Devices
} // namespace Service::Nvidia

View file

@ -10,8 +10,9 @@
#include "video_core/renderer_base.h"
namespace Service::Nvidia::Devices {
nvhost_vic::nvhost_vic(Core::System& system, std::shared_ptr<nvmap> nvmap_dev)
: nvhost_nvdec_common(system, std::move(nvmap_dev)) {
nvhost_vic::nvhost_vic(Core::System& system, std::shared_ptr<nvmap> nvmap_dev,
SyncpointManager& syncpoint_manager)
: nvhost_nvdec_common(system, std::move(nvmap_dev), syncpoint_manager) {
vic_device = true;
}
nvhost_vic::~nvhost_vic() = default;

View file

@ -7,11 +7,11 @@
#include "core/hle/service/nvdrv/devices/nvhost_nvdec_common.h"
namespace Service::Nvidia::Devices {
class nvmap;
class nvhost_vic final : public nvhost_nvdec_common {
public:
explicit nvhost_vic(Core::System& system, std::shared_ptr<nvmap> nvmap_dev);
explicit nvhost_vic(Core::System& system, std::shared_ptr<nvmap> nvmap_dev,
SyncpointManager& syncpoint_manager);
~nvhost_vic();
NvResult Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;

View file

@ -55,9 +55,11 @@ Module::Module(Core::System& system) : syncpoint_manager{system.GPU()} {
devices["/dev/nvdisp_disp0"] = std::make_shared<Devices::nvdisp_disp0>(system, nvmap_dev);
devices["/dev/nvhost-ctrl"] =
std::make_shared<Devices::nvhost_ctrl>(system, events_interface, syncpoint_manager);
devices["/dev/nvhost-nvdec"] = std::make_shared<Devices::nvhost_nvdec>(system, nvmap_dev);
devices["/dev/nvhost-nvdec"] =
std::make_shared<Devices::nvhost_nvdec>(system, nvmap_dev, syncpoint_manager);
devices["/dev/nvhost-nvjpg"] = std::make_shared<Devices::nvhost_nvjpg>(system);
devices["/dev/nvhost-vic"] = std::make_shared<Devices::nvhost_vic>(system, nvmap_dev);
devices["/dev/nvhost-vic"] =
std::make_shared<Devices::nvhost_vic>(system, nvmap_dev, syncpoint_manager);
}
Module::~Module() = default;

View file

@ -246,7 +246,6 @@ add_library(video_core STATIC
texture_cache/samples_helper.h
texture_cache/slot_vector.h
texture_cache/texture_cache.h
texture_cache/texture_cache.inl
texture_cache/types.h
texture_cache/util.cpp
texture_cache/util.h

View file

@ -33,8 +33,7 @@ CDmaPusher::CDmaPusher(GPU& gpu_)
: gpu{gpu_}, nvdec_processor(std::make_shared<Nvdec>(gpu)),
vic_processor(std::make_unique<Vic>(gpu, nvdec_processor)),
host1x_processor(std::make_unique<Host1x>(gpu)),
nvdec_sync(std::make_unique<SyncptIncrManager>(gpu)),
vic_sync(std::make_unique<SyncptIncrManager>(gpu)) {}
sync_manager(std::make_unique<SyncptIncrManager>(gpu)) {}
CDmaPusher::~CDmaPusher() = default;
@ -110,10 +109,10 @@ void CDmaPusher::ExecuteCommand(u32 state_offset, u32 data) {
const auto syncpoint_id = static_cast<u32>(data & 0xFF);
const auto cond = static_cast<u32>((data >> 8) & 0xFF);
if (cond == 0) {
nvdec_sync->Increment(syncpoint_id);
sync_manager->Increment(syncpoint_id);
} else {
nvdec_sync->IncrementWhenDone(static_cast<u32>(current_class), syncpoint_id);
nvdec_sync->SignalDone(syncpoint_id);
sync_manager->SignalDone(
sync_manager->IncrementWhenDone(static_cast<u32>(current_class), syncpoint_id));
}
break;
}
@ -135,10 +134,10 @@ void CDmaPusher::ExecuteCommand(u32 state_offset, u32 data) {
const auto syncpoint_id = static_cast<u32>(data & 0xFF);
const auto cond = static_cast<u32>((data >> 8) & 0xFF);
if (cond == 0) {
vic_sync->Increment(syncpoint_id);
sync_manager->Increment(syncpoint_id);
} else {
vic_sync->IncrementWhenDone(static_cast<u32>(current_class), syncpoint_id);
vic_sync->SignalDone(syncpoint_id);
sync_manager->SignalDone(
sync_manager->IncrementWhenDone(static_cast<u32>(current_class), syncpoint_id));
}
break;
}

View file

@ -116,12 +116,10 @@ private:
void ThiStateWrite(ThiRegisters& state, u32 state_offset, const std::vector<u32>& arguments);
GPU& gpu;
std::shared_ptr<Nvdec> nvdec_processor;
std::unique_ptr<Vic> vic_processor;
std::unique_ptr<Host1x> host1x_processor;
std::unique_ptr<SyncptIncrManager> nvdec_sync;
std::unique_ptr<SyncptIncrManager> vic_sync;
std::shared_ptr<Tegra::Nvdec> nvdec_processor;
std::unique_ptr<Tegra::Vic> vic_processor;
std::unique_ptr<Tegra::Host1x> host1x_processor;
std::unique_ptr<SyncptIncrManager> sync_manager;
ChClassId current_class{};
ThiRegisters vic_thi_state{};
ThiRegisters nvdec_thi_state{};

View file

@ -10,22 +10,14 @@ Tegra::Host1x::Host1x(GPU& gpu_) : gpu(gpu_) {}
Tegra::Host1x::~Host1x() = default;
void Tegra::Host1x::StateWrite(u32 offset, u32 arguments) {
u8* const state_offset = reinterpret_cast<u8*>(&state) + offset * sizeof(u32);
std::memcpy(state_offset, &arguments, sizeof(u32));
}
void Tegra::Host1x::ProcessMethod(Method method, const std::vector<u32>& arguments) {
StateWrite(static_cast<u32>(method), arguments[0]);
void Tegra::Host1x::ProcessMethod(Method method, u32 argument) {
switch (method) {
case Method::WaitSyncpt:
Execute(arguments[0]);
break;
case Method::LoadSyncptPayload32:
syncpoint_value = arguments[0];
syncpoint_value = argument;
break;
case Method::WaitSyncpt:
case Method::WaitSyncpt32:
Execute(arguments[0]);
Execute(argument);
break;
default:
UNIMPLEMENTED_MSG("Host1x method 0x{:X}", static_cast<u32>(method));
@ -34,6 +26,5 @@ void Tegra::Host1x::ProcessMethod(Method method, const std::vector<u32>& argumen
}
void Tegra::Host1x::Execute(u32 data) {
// This method waits on a valid syncpoint.
// TODO: Implement when proper Async is in place
gpu.WaitFence(data, syncpoint_value);
}

View file

@ -14,64 +14,23 @@ class Nvdec;
class Host1x {
public:
struct Host1xClassRegisters {
u32 incr_syncpt{};
u32 incr_syncpt_ctrl{};
u32 incr_syncpt_error{};
INSERT_PADDING_WORDS(5);
u32 wait_syncpt{};
u32 wait_syncpt_base{};
u32 wait_syncpt_incr{};
u32 load_syncpt_base{};
u32 incr_syncpt_base{};
u32 clear{};
u32 wait{};
u32 wait_with_interrupt{};
u32 delay_use{};
u32 tick_count_high{};
u32 tick_count_low{};
u32 tick_ctrl{};
INSERT_PADDING_WORDS(23);
u32 ind_ctrl{};
u32 ind_off2{};
u32 ind_off{};
std::array<u32, 31> ind_data{};
INSERT_PADDING_WORDS(1);
u32 load_syncpoint_payload32{};
u32 stall_ctrl{};
u32 wait_syncpt32{};
u32 wait_syncpt_base32{};
u32 load_syncpt_base32{};
u32 incr_syncpt_base32{};
u32 stall_count_high{};
u32 stall_count_low{};
u32 xref_ctrl{};
u32 channel_xref_high{};
u32 channel_xref_low{};
};
static_assert(sizeof(Host1xClassRegisters) == 0x164, "Host1xClassRegisters is an invalid size");
enum class Method : u32 {
WaitSyncpt = offsetof(Host1xClassRegisters, wait_syncpt) / 4,
LoadSyncptPayload32 = offsetof(Host1xClassRegisters, load_syncpoint_payload32) / 4,
WaitSyncpt32 = offsetof(Host1xClassRegisters, wait_syncpt32) / 4,
WaitSyncpt = 0x8,
LoadSyncptPayload32 = 0x4e,
WaitSyncpt32 = 0x50,
};
explicit Host1x(GPU& gpu);
~Host1x();
/// Writes the method into the state, Invoke Execute() if encountered
void ProcessMethod(Method method, const std::vector<u32>& arguments);
void ProcessMethod(Method method, u32 argument);
private:
/// For Host1x, execute is waiting on a syncpoint previously written into the state
void Execute(u32 data);
/// Write argument into the provided offset
void StateWrite(u32 offset, u32 arguments);
u32 syncpoint_value{};
Host1xClassRegisters state{};
GPU& gpu;
};

View file

@ -11,6 +11,7 @@
namespace VideoCommon {
/// Container to push objects to be destroyed a few ticks in the future
template <typename T, size_t TICKS_TO_DESTROY>
class DelayedDestructionRing {
public:

View file

@ -1,4 +1,4 @@
// Copyright 2019 yuzu Emulator Project
// Copyright 2020 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.

View file

@ -734,7 +734,6 @@ void RasterizerOpenGL::InvalidateTextureCache(VAddr addr, u64 size) {
if (addr == 0 || size == 0) {
return;
}
auto lock = texture_cache.AcquireLock();
texture_cache.UnmapMemory(addr, size);
}

View file

@ -354,7 +354,7 @@ void ApplySwizzle(GLuint handle, PixelFormat format, std::array<SwizzleSource, 4
};
default:
UNIMPLEMENTED_MSG("Unimplemented copy target={}", target);
return CopyOrigin{.x = 0, .y = 0, .z = 0};
return CopyOrigin{.level = 0, .x = 0, .y = 0, .z = 0};
}
}

View file

@ -114,7 +114,6 @@ void UtilShaders::BlockLinearUpload3D(Image& image, const ImageBufferMap& map, s
const GLenum store_format = StoreFormat(BytesPerBlock(image.info.format));
for (const SwizzleParameters& swizzle : swizzles) {
const Extent3D block = swizzle.block;
const Extent3D num_tiles = swizzle.num_tiles;
const size_t input_offset = swizzle.buffer_offset + buffer_offset;

View file

@ -657,7 +657,7 @@ void RasterizerVulkan::FlushRegion(VAddr addr, u64 size) {
query_cache.FlushRegion(addr, size);
}
void RasterizerVulkan::InvalidateExceptTextureCache(VAddr addr, u64 size) {
void Vulkan::RasterizerVulkan::InvalidateExceptTextureCache(VAddr addr, u64 size) {
if (addr == 0 || size == 0) {
return;
}
@ -666,11 +666,10 @@ void RasterizerVulkan::InvalidateExceptTextureCache(VAddr addr, u64 size) {
query_cache.InvalidateRegion(addr, size);
}
void RasterizerVulkan::InvalidateTextureCache(VAddr addr, u64 size) {
void Vulkan::RasterizerVulkan::InvalidateTextureCache(VAddr addr, u64 size) {
if (addr == 0 || size == 0) {
return;
}
auto lock = texture_cache.AcquireLock();
texture_cache.UnmapMemory(addr, size);
}

View file

@ -51,11 +51,6 @@ struct VKScreenInfo;
class StateTracker;
class BufferBindings;
struct Transition {
ImageView* view;
VkImageLayout* layout;
};
class RasterizerVulkan final : public VideoCore::RasterizerAccelerated {
public:
explicit RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra::GPU& gpu_,

View file

@ -6,6 +6,10 @@
#include <string>
#include "video_core/texture_cache/formatter.h"
#include "video_core/texture_cache/image_base.h"
#include "video_core/texture_cache/image_info.h"
#include "video_core/texture_cache/image_view_base.h"
#include "video_core/texture_cache/render_targets.h"
namespace VideoCommon {

View file

@ -8,7 +8,8 @@
#include <fmt/format.h>
#include "video_core/texture_cache/texture_cache.h"
#include "video_core/surface.h"
#include "video_core/texture_cache/types.h"
template <>
struct fmt::formatter<VideoCore::Surface::PixelFormat> : fmt::formatter<fmt::string_view> {

File diff suppressed because it is too large Load diff

View file

@ -40,10 +40,12 @@
#include "common/div_ceil.h"
#include "video_core/compatible_formats.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/memory_manager.h"
#include "video_core/surface.h"
#include "video_core/texture_cache/decode_bc4.h"
#include "video_core/texture_cache/format_lookup_table.h"
#include "video_core/texture_cache/formatter.h"
#include "video_core/texture_cache/samples_helper.h"
#include "video_core/texture_cache/util.h"
#include "video_core/textures/astc.h"
#include "video_core/textures/decoders.h"

View file

@ -292,12 +292,48 @@ GMainWindow::GMainWindow()
connect(&mouse_hide_timer, &QTimer::timeout, this, &GMainWindow::HideMouseCursor);
connect(ui.menubar, &QMenuBar::hovered, this, &GMainWindow::ShowMouseCursor);
MigrateConfigFiles();
ui.action_Fullscreen->setChecked(false);
QStringList args = QApplication::arguments();
if (args.length() >= 2) {
BootGame(args[1]);
if (args.size() < 2) {
return;
}
MigrateConfigFiles();
QString game_path;
for (int i = 1; i < args.size(); ++i) {
// Preserves drag/drop functionality
if (args.size() == 2 && !args[1].startsWith(QChar::fromLatin1('-'))) {
game_path = args[1];
break;
}
// Launch game in fullscreen mode
if (args[i] == QStringLiteral("-f")) {
ui.action_Fullscreen->setChecked(true);
continue;
}
// Launch game at path
if (args[i] == QStringLiteral("-g")) {
if (i >= args.size() - 1) {
continue;
}
if (args[i + 1].startsWith(QChar::fromLatin1('-'))) {
continue;
}
game_path = args[++i];
}
}
if (!game_path.isEmpty()) {
BootGame(game_path);
}
}
GMainWindow::~GMainWindow() {