summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat
-rw-r--r--src/core/hle/kernel/k_thread.h10
-rw-r--r--src/core/hle/kernel/svc/svc_ipc.cpp37
-rw-r--r--src/core/hle/kernel/svc/svc_synchronization.cpp41
-rw-r--r--src/video_core/compatible_formats.cpp6
-rw-r--r--src/video_core/renderer_vulkan/vk_texture_cache.cpp45
-rw-r--r--src/video_core/renderer_vulkan/vk_texture_cache.h5
-rw-r--r--src/video_core/texture_cache/types.h1
-rw-r--r--src/video_core/texture_cache/util.cpp9
-rw-r--r--src/video_core/vulkan_common/vulkan_device.h6
9 files changed, 110 insertions, 50 deletions
diff --git a/src/core/hle/kernel/k_thread.h b/src/core/hle/kernel/k_thread.h
index dd662b3f8..d178c2453 100644
--- a/src/core/hle/kernel/k_thread.h
+++ b/src/core/hle/kernel/k_thread.h
@@ -338,6 +338,15 @@ public:
return m_parent != nullptr;
}
+ std::span<KSynchronizationObject*> GetSynchronizationObjectBuffer() {
+ return m_sync_object_buffer.sync_objects;
+ }
+
+ std::span<Handle> GetHandleBuffer() {
+ return {m_sync_object_buffer.handles.data() + Svc::ArgumentHandleCountMax,
+ Svc::ArgumentHandleCountMax};
+ }
+
u16 GetUserDisableCount() const;
void SetInterruptFlag();
void ClearInterruptFlag();
@@ -855,6 +864,7 @@ private:
u32* m_light_ipc_data{};
KProcessAddress m_tls_address{};
KLightLock m_activity_pause_lock;
+ SyncObjectBuffer m_sync_object_buffer{};
s64 m_schedule_count{};
s64 m_last_scheduled_tick{};
std::array<QueueEntry, Core::Hardware::NUM_CPU_CORES> m_per_core_priority_queue_entry{};
diff --git a/src/core/hle/kernel/svc/svc_ipc.cpp b/src/core/hle/kernel/svc/svc_ipc.cpp
index 60247df2e..bb94f6934 100644
--- a/src/core/hle/kernel/svc/svc_ipc.cpp
+++ b/src/core/hle/kernel/svc/svc_ipc.cpp
@@ -38,22 +38,31 @@ Result SendAsyncRequestWithUserBuffer(Core::System& system, Handle* out_event_ha
Result ReplyAndReceive(Core::System& system, s32* out_index, uint64_t handles_addr, s32 num_handles,
Handle reply_target, s64 timeout_ns) {
+ // Ensure number of handles is valid.
+ R_UNLESS(0 <= num_handles && num_handles <= ArgumentHandleCountMax, ResultOutOfRange);
+
+ // Get the synchronization context.
auto& kernel = system.Kernel();
auto& handle_table = GetCurrentProcess(kernel).GetHandleTable();
-
- R_UNLESS(0 <= num_handles && num_handles <= ArgumentHandleCountMax, ResultOutOfRange);
- R_UNLESS(GetCurrentMemory(kernel).IsValidVirtualAddressRange(
- handles_addr, static_cast<u64>(sizeof(Handle) * num_handles)),
- ResultInvalidPointer);
-
- std::array<Handle, Svc::ArgumentHandleCountMax> handles;
- GetCurrentMemory(kernel).ReadBlock(handles_addr, handles.data(), sizeof(Handle) * num_handles);
-
- // Convert handle list to object table.
- std::array<KSynchronizationObject*, Svc::ArgumentHandleCountMax> objs;
- R_UNLESS(handle_table.GetMultipleObjects<KSynchronizationObject>(objs.data(), handles.data(),
- num_handles),
- ResultInvalidHandle);
+ auto objs = GetCurrentThread(kernel).GetSynchronizationObjectBuffer();
+ auto handles = GetCurrentThread(kernel).GetHandleBuffer();
+
+ // Copy user handles.
+ if (num_handles > 0) {
+ // Ensure we can try to get the handles.
+ R_UNLESS(GetCurrentMemory(kernel).IsValidVirtualAddressRange(
+ handles_addr, static_cast<u64>(sizeof(Handle) * num_handles)),
+ ResultInvalidPointer);
+
+ // Get the handles.
+ GetCurrentMemory(kernel).ReadBlock(handles_addr, handles.data(),
+ sizeof(Handle) * num_handles);
+
+ // Convert the handles to objects.
+ R_UNLESS(handle_table.GetMultipleObjects<KSynchronizationObject>(
+ objs.data(), handles.data(), num_handles),
+ ResultInvalidHandle);
+ }
// Ensure handles are closed when we're done.
SCOPE_EXIT({
diff --git a/src/core/hle/kernel/svc/svc_synchronization.cpp b/src/core/hle/kernel/svc/svc_synchronization.cpp
index 53df5bcd8..f02d03f30 100644
--- a/src/core/hle/kernel/svc/svc_synchronization.cpp
+++ b/src/core/hle/kernel/svc/svc_synchronization.cpp
@@ -47,21 +47,35 @@ Result ResetSignal(Core::System& system, Handle handle) {
R_THROW(ResultInvalidHandle);
}
-static Result WaitSynchronization(Core::System& system, int32_t* out_index, const Handle* handles,
- int32_t num_handles, int64_t timeout_ns) {
+/// Wait for the given handles to synchronize, timeout after the specified nanoseconds
+Result WaitSynchronization(Core::System& system, int32_t* out_index, u64 user_handles,
+ int32_t num_handles, int64_t timeout_ns) {
+ LOG_TRACE(Kernel_SVC, "called user_handles={:#x}, num_handles={}, timeout_ns={}", user_handles,
+ num_handles, timeout_ns);
+
// Ensure number of handles is valid.
R_UNLESS(0 <= num_handles && num_handles <= Svc::ArgumentHandleCountMax, ResultOutOfRange);
// Get the synchronization context.
auto& kernel = system.Kernel();
auto& handle_table = GetCurrentProcess(kernel).GetHandleTable();
- std::array<KSynchronizationObject*, Svc::ArgumentHandleCountMax> objs;
+ auto objs = GetCurrentThread(kernel).GetSynchronizationObjectBuffer();
+ auto handles = GetCurrentThread(kernel).GetHandleBuffer();
// Copy user handles.
if (num_handles > 0) {
+ // Ensure we can try to get the handles.
+ R_UNLESS(GetCurrentMemory(kernel).IsValidVirtualAddressRange(
+ user_handles, static_cast<u64>(sizeof(Handle) * num_handles)),
+ ResultInvalidPointer);
+
+ // Get the handles.
+ GetCurrentMemory(kernel).ReadBlock(user_handles, handles.data(),
+ sizeof(Handle) * num_handles);
+
// Convert the handles to objects.
- R_UNLESS(handle_table.GetMultipleObjects<KSynchronizationObject>(objs.data(), handles,
- num_handles),
+ R_UNLESS(handle_table.GetMultipleObjects<KSynchronizationObject>(
+ objs.data(), handles.data(), num_handles),
ResultInvalidHandle);
}
@@ -80,23 +94,6 @@ static Result WaitSynchronization(Core::System& system, int32_t* out_index, cons
R_RETURN(res);
}
-/// Wait for the given handles to synchronize, timeout after the specified nanoseconds
-Result WaitSynchronization(Core::System& system, int32_t* out_index, u64 user_handles,
- int32_t num_handles, int64_t timeout_ns) {
- LOG_TRACE(Kernel_SVC, "called user_handles={:#x}, num_handles={}, timeout_ns={}", user_handles,
- num_handles, timeout_ns);
-
- // Ensure number of handles is valid.
- R_UNLESS(0 <= num_handles && num_handles <= Svc::ArgumentHandleCountMax, ResultOutOfRange);
- std::array<Handle, Svc::ArgumentHandleCountMax> handles;
- if (num_handles > 0) {
- GetCurrentMemory(system.Kernel())
- .ReadBlock(user_handles, handles.data(), num_handles * sizeof(Handle));
- }
-
- R_RETURN(WaitSynchronization(system, out_index, handles.data(), num_handles, timeout_ns));
-}
-
/// Resumes a thread waiting on WaitSynchronization
Result CancelSynchronization(Core::System& system, Handle handle) {
LOG_TRACE(Kernel_SVC, "called handle=0x{:X}", handle);
diff --git a/src/video_core/compatible_formats.cpp b/src/video_core/compatible_formats.cpp
index ab4f4d407..87d69ebc5 100644
--- a/src/video_core/compatible_formats.cpp
+++ b/src/video_core/compatible_formats.cpp
@@ -272,6 +272,9 @@ constexpr Table MakeNonNativeBgrCopyTable() {
bool IsViewCompatible(PixelFormat format_a, PixelFormat format_b, bool broken_views,
bool native_bgr) {
+ if (format_a == format_b) {
+ return true;
+ }
if (broken_views) {
// If format views are broken, only accept formats that are identical.
return format_a == format_b;
@@ -282,6 +285,9 @@ bool IsViewCompatible(PixelFormat format_a, PixelFormat format_b, bool broken_vi
}
bool IsCopyCompatible(PixelFormat format_a, PixelFormat format_b, bool native_bgr) {
+ if (format_a == format_b) {
+ return true;
+ }
static constexpr Table BGR_TABLE = MakeNativeBgrCopyTable();
static constexpr Table NO_BGR_TABLE = MakeNonNativeBgrCopyTable();
return IsSupported(native_bgr ? BGR_TABLE : NO_BGR_TABLE, format_a, format_b);
diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
index 8385b5509..3aac3cfab 100644
--- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
@@ -36,8 +36,10 @@ using VideoCommon::ImageFlagBits;
using VideoCommon::ImageInfo;
using VideoCommon::ImageType;
using VideoCommon::SubresourceRange;
+using VideoCore::Surface::BytesPerBlock;
using VideoCore::Surface::IsPixelFormatASTC;
using VideoCore::Surface::IsPixelFormatInteger;
+using VideoCore::Surface::SurfaceType;
namespace {
constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
@@ -130,7 +132,7 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
[[nodiscard]] VkImageCreateInfo MakeImageCreateInfo(const Device& device, const ImageInfo& info) {
const PixelFormat format = StorageFormat(info.format);
const auto format_info = MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, false, format);
- VkImageCreateFlags flags = VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT;
+ VkImageCreateFlags flags{};
if (info.type == ImageType::e2D && info.resources.layers >= 6 &&
info.size.width == info.size.height && !device.HasBrokenCubeImageCompability()) {
flags |= VK_IMAGE_CREATE_CUBE_COMPATIBLE_BIT;
@@ -163,11 +165,24 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
}
[[nodiscard]] vk::Image MakeImage(const Device& device, const MemoryAllocator& allocator,
- const ImageInfo& info) {
+ const ImageInfo& info, std::span<const VkFormat> view_formats) {
if (info.type == ImageType::Buffer) {
return vk::Image{};
}
- return allocator.CreateImage(MakeImageCreateInfo(device, info));
+ VkImageCreateInfo image_ci = MakeImageCreateInfo(device, info);
+ const VkImageFormatListCreateInfo image_format_list = {
+ .sType = VK_STRUCTURE_TYPE_IMAGE_FORMAT_LIST_CREATE_INFO,
+ .pNext = nullptr,
+ .viewFormatCount = static_cast<u32>(view_formats.size()),
+ .pViewFormats = view_formats.data(),
+ };
+ if (view_formats.size() > 1) {
+ image_ci.flags |= VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT;
+ if (device.IsKhrImageFormatListSupported()) {
+ image_ci.pNext = &image_format_list;
+ }
+ }
+ return allocator.CreateImage(image_ci);
}
[[nodiscard]] VkImageAspectFlags ImageAspectMask(PixelFormat format) {
@@ -806,6 +821,23 @@ TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& sched
astc_decoder_pass.emplace(device, scheduler, descriptor_pool, staging_buffer_pool,
compute_pass_descriptor_queue, memory_allocator);
}
+ if (!device.IsKhrImageFormatListSupported()) {
+ return;
+ }
+ for (size_t index_a = 0; index_a < VideoCore::Surface::MaxPixelFormat; index_a++) {
+ const auto image_format = static_cast<PixelFormat>(index_a);
+ if (IsPixelFormatASTC(image_format) && !device.IsOptimalAstcSupported()) {
+ view_formats[index_a].push_back(VK_FORMAT_A8B8G8R8_UNORM_PACK32);
+ }
+ for (size_t index_b = 0; index_b < VideoCore::Surface::MaxPixelFormat; index_b++) {
+ const auto view_format = static_cast<PixelFormat>(index_b);
+ if (VideoCore::Surface::IsViewCompatible(image_format, view_format, false, true)) {
+ const auto view_info =
+ MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, true, view_format);
+ view_formats[index_a].push_back(view_info.format);
+ }
+ }
+ }
}
void TextureCacheRuntime::Finish() {
@@ -1265,8 +1297,8 @@ void TextureCacheRuntime::TickFrame() {}
Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu_addr_,
VAddr cpu_addr_)
: VideoCommon::ImageBase(info_, gpu_addr_, cpu_addr_), scheduler{&runtime_.scheduler},
- runtime{&runtime_},
- original_image(MakeImage(runtime_.device, runtime_.memory_allocator, info)),
+ runtime{&runtime_}, original_image(MakeImage(runtime_.device, runtime_.memory_allocator, info,
+ runtime->ViewFormats(info.format))),
aspect_mask(ImageAspectMask(info.format)) {
if (IsPixelFormatASTC(info.format) && !runtime->device.IsOptimalAstcSupported()) {
if (Settings::values.async_astc.GetValue()) {
@@ -1471,7 +1503,8 @@ bool Image::ScaleUp(bool ignore) {
auto scaled_info = info;
scaled_info.size.width = scaled_width;
scaled_info.size.height = scaled_height;
- scaled_image = MakeImage(runtime->device, runtime->memory_allocator, scaled_info);
+ scaled_image = MakeImage(runtime->device, runtime->memory_allocator, scaled_info,
+ runtime->ViewFormats(info.format));
ignore = false;
}
current_image = *scaled_image;
diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.h b/src/video_core/renderer_vulkan/vk_texture_cache.h
index 220943116..6621210ea 100644
--- a/src/video_core/renderer_vulkan/vk_texture_cache.h
+++ b/src/video_core/renderer_vulkan/vk_texture_cache.h
@@ -103,6 +103,10 @@ public:
[[nodiscard]] VkBuffer GetTemporaryBuffer(size_t needed_size);
+ std::span<const VkFormat> ViewFormats(PixelFormat format) {
+ return view_formats[static_cast<std::size_t>(format)];
+ }
+
void BarrierFeedbackLoop();
const Device& device;
@@ -113,6 +117,7 @@ public:
RenderPassCache& render_pass_cache;
std::optional<ASTCDecoderPass> astc_decoder_pass;
const Settings::ResolutionScalingInfo& resolution;
+ std::array<std::vector<VkFormat>, VideoCore::Surface::MaxPixelFormat> view_formats;
static constexpr size_t indexing_slots = 8 * sizeof(size_t);
std::array<vk::Buffer, indexing_slots> buffers{};
diff --git a/src/video_core/texture_cache/types.h b/src/video_core/texture_cache/types.h
index a0e10643f..0453456b4 100644
--- a/src/video_core/texture_cache/types.h
+++ b/src/video_core/texture_cache/types.h
@@ -54,7 +54,6 @@ enum class RelaxedOptions : u32 {
Format = 1 << 1,
Samples = 1 << 2,
ForceBrokenViews = 1 << 3,
- FormatBpp = 1 << 4,
};
DECLARE_ENUM_FLAG_OPERATORS(RelaxedOptions)
diff --git a/src/video_core/texture_cache/util.cpp b/src/video_core/texture_cache/util.cpp
index 9a618a57a..0de6ed09d 100644
--- a/src/video_core/texture_cache/util.cpp
+++ b/src/video_core/texture_cache/util.cpp
@@ -1201,8 +1201,7 @@ std::optional<SubresourceBase> FindSubresource(const ImageInfo& candidate, const
// Format checking is relaxed, but we still have to check for matching bytes per block.
// This avoids creating a view for blits on UE4 titles where formats with different bytes
// per block are aliased.
- if (BytesPerBlock(existing.format) != BytesPerBlock(candidate.format) &&
- False(options & RelaxedOptions::FormatBpp)) {
+ if (BytesPerBlock(existing.format) != BytesPerBlock(candidate.format)) {
return std::nullopt;
}
} else {
@@ -1233,11 +1232,7 @@ std::optional<SubresourceBase> FindSubresource(const ImageInfo& candidate, const
}
const bool strict_size = False(options & RelaxedOptions::Size);
if (!IsBlockLinearSizeCompatible(existing, candidate, base->level, 0, strict_size)) {
- if (False(options & RelaxedOptions::FormatBpp)) {
- return std::nullopt;
- } else if (!IsBlockLinearSizeCompatibleBPPRelaxed(existing, candidate, base->level, 0)) {
- return std::nullopt;
- }
+ return std::nullopt;
}
// TODO: compare block sizes
return base;
diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h
index 1f17265d5..3ace1fb03 100644
--- a/src/video_core/vulkan_common/vulkan_device.h
+++ b/src/video_core/vulkan_common/vulkan_device.h
@@ -77,6 +77,7 @@ VK_DEFINE_HANDLE(VmaAllocator)
EXTENSION(KHR, SPIRV_1_4, spirv_1_4) \
EXTENSION(KHR, SWAPCHAIN, swapchain) \
EXTENSION(KHR, SWAPCHAIN_MUTABLE_FORMAT, swapchain_mutable_format) \
+ EXTENSION(KHR, IMAGE_FORMAT_LIST, image_format_list) \
EXTENSION(NV, DEVICE_DIAGNOSTICS_CONFIG, device_diagnostics_config) \
EXTENSION(NV, GEOMETRY_SHADER_PASSTHROUGH, geometry_shader_passthrough) \
EXTENSION(NV, VIEWPORT_ARRAY2, viewport_array2) \
@@ -408,6 +409,11 @@ public:
return extensions.workgroup_memory_explicit_layout;
}
+ /// Returns true if the device supports VK_KHR_image_format_list.
+ bool IsKhrImageFormatListSupported() const {
+ return extensions.image_format_list || instance_version >= VK_API_VERSION_1_2;
+ }
+
/// Returns true if the device supports VK_EXT_primitive_topology_list_restart.
bool IsTopologyListPrimitiveRestartSupported() const {
return features.primitive_topology_list_restart.primitiveTopologyListRestart;
generated by cgit v1.2.3 (git 2.25.1) at 2024-06-02 17:35:42 +0200