summaryrefslogtreecommitdiffstats
path: root/src/video_core
diff options
context:
space:
mode:
Diffstat (limited to 'src/video_core')
-rw-r--r--src/video_core/gpu.cpp5
-rw-r--r--src/video_core/renderer_opengl/gl_device.cpp10
-rw-r--r--src/video_core/renderer_opengl/gl_device.h8
-rw-r--r--src/video_core/renderer_opengl/gl_shader_cache.cpp76
-rw-r--r--src/video_core/renderer_opengl/gl_shader_cache.h1
-rw-r--r--src/video_core/renderer_opengl/renderer_opengl.cpp4
-rw-r--r--src/video_core/renderer_vulkan/renderer_vulkan.cpp14
-rw-r--r--src/video_core/renderer_vulkan/vk_buffer_cache.cpp3
-rw-r--r--src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp10
-rw-r--r--src/video_core/renderer_vulkan/vk_pipeline_cache.cpp9
-rw-r--r--src/video_core/renderer_vulkan/vk_swapchain.cpp15
-rw-r--r--src/video_core/renderer_vulkan/vk_swapchain.h14
-rw-r--r--src/video_core/vulkan_common/vulkan_device.cpp7
-rw-r--r--src/video_core/vulkan_common/vulkan_device.h10
14 files changed, 128 insertions, 58 deletions
diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp
index 28b38273e..c6d54be63 100644
--- a/src/video_core/gpu.cpp
+++ b/src/video_core/gpu.cpp
@@ -223,8 +223,6 @@ struct GPU::Impl {
/// core timing events.
void Start() {
gpu_thread.StartThread(*renderer, renderer->Context(), *scheduler);
- cpu_context = renderer->GetRenderWindow().CreateSharedContext();
- cpu_context->MakeCurrent();
}
void NotifyShutdown() {
@@ -235,6 +233,9 @@ struct GPU::Impl {
/// Obtain the CPU Context
void ObtainContext() {
+ if (!cpu_context) {
+ cpu_context = renderer->GetRenderWindow().CreateSharedContext();
+ }
cpu_context->MakeCurrent();
}
diff --git a/src/video_core/renderer_opengl/gl_device.cpp b/src/video_core/renderer_opengl/gl_device.cpp
index e2e3dac34..cee5c3247 100644
--- a/src/video_core/renderer_opengl/gl_device.cpp
+++ b/src/video_core/renderer_opengl/gl_device.cpp
@@ -112,7 +112,7 @@ bool IsASTCSupported() {
}
} // Anonymous namespace
-Device::Device() {
+Device::Device(Core::Frontend::EmuWindow& emu_window) {
if (!GLAD_GL_VERSION_4_6) {
LOG_ERROR(Render_OpenGL, "OpenGL 4.6 is not available");
throw std::runtime_error{"Insufficient version"};
@@ -126,9 +126,9 @@ Device::Device() {
const bool is_intel = vendor_name == "Intel";
#ifdef __unix__
- const bool is_linux = true;
+ constexpr bool is_linux = true;
#else
- const bool is_linux = false;
+ constexpr bool is_linux = false;
#endif
bool disable_fast_buffer_sub_data = false;
@@ -193,9 +193,11 @@ Device::Device() {
}
}
+ strict_context_required = emu_window.StrictContextRequired();
// Blocks AMD and Intel OpenGL drivers on Windows from using asynchronous shader compilation.
+ // Blocks EGL on Wayland from using asynchronous shader compilation.
use_asynchronous_shaders = Settings::values.use_asynchronous_shaders.GetValue() &&
- !(is_amd || (is_intel && !is_linux));
+ !(is_amd || (is_intel && !is_linux)) && !strict_context_required;
use_driver_cache = is_nvidia;
LOG_INFO(Render_OpenGL, "Renderer_VariableAOFFI: {}", has_variable_aoffi);
diff --git a/src/video_core/renderer_opengl/gl_device.h b/src/video_core/renderer_opengl/gl_device.h
index 5ef51ebcf..2a72d84be 100644
--- a/src/video_core/renderer_opengl/gl_device.h
+++ b/src/video_core/renderer_opengl/gl_device.h
@@ -5,6 +5,7 @@
#include <cstddef>
#include "common/common_types.h"
+#include "core/frontend/emu_window.h"
#include "shader_recompiler/stage.h"
namespace Settings {
@@ -15,7 +16,7 @@ namespace OpenGL {
class Device {
public:
- explicit Device();
+ explicit Device(Core::Frontend::EmuWindow& emu_window);
[[nodiscard]] std::string GetVendorName() const;
@@ -173,6 +174,10 @@ public:
return can_report_memory;
}
+ bool StrictContextRequired() const {
+ return strict_context_required;
+ }
+
private:
static bool TestVariableAoffi();
static bool TestPreciseBug();
@@ -216,6 +221,7 @@ private:
bool has_cbuf_ftou_bug{};
bool has_bool_ref_bug{};
bool can_report_memory{};
+ bool strict_context_required{};
std::string vendor_name;
};
diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp
index a59d0d24e..fff55d585 100644
--- a/src/video_core/renderer_opengl/gl_shader_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp
@@ -174,6 +174,7 @@ ShaderCache::ShaderCache(RasterizerOpenGL& rasterizer_, Core::Frontend::EmuWindo
texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, program_manager{program_manager_},
state_tracker{state_tracker_}, shader_notify{shader_notify_},
use_asynchronous_shaders{device.UseAsynchronousShaders()},
+ strict_context_required{device.StrictContextRequired()},
profile{
.supported_spirv = 0x00010000,
@@ -255,9 +256,14 @@ void ShaderCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading,
}
shader_cache_filename = base_dir / "opengl.bin";
- if (!workers) {
+ if (!workers && !strict_context_required) {
workers = CreateWorkers();
}
+ std::optional<Context> strict_context;
+ if (strict_context_required) {
+ strict_context.emplace(emu_window);
+ }
+
struct {
std::mutex mutex;
size_t total{};
@@ -265,44 +271,49 @@ void ShaderCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading,
bool has_loaded{};
} state;
+ const auto queue_work{[&](Common::UniqueFunction<void, Context*>&& work) {
+ if (strict_context_required) {
+ work(&strict_context.value());
+ } else {
+ workers->QueueWork(std::move(work));
+ }
+ }};
const auto load_compute{[&](std::ifstream& file, FileEnvironment env) {
ComputePipelineKey key;
file.read(reinterpret_cast<char*>(&key), sizeof(key));
- workers->QueueWork(
- [this, key, env = std::move(env), &state, &callback](Context* ctx) mutable {
- ctx->pools.ReleaseContents();
- auto pipeline{CreateComputePipeline(ctx->pools, key, env)};
- std::scoped_lock lock{state.mutex};
- if (pipeline) {
- compute_cache.emplace(key, std::move(pipeline));
- }
- ++state.built;
- if (state.has_loaded) {
- callback(VideoCore::LoadCallbackStage::Build, state.built, state.total);
- }
- });
+ queue_work([this, key, env = std::move(env), &state, &callback](Context* ctx) mutable {
+ ctx->pools.ReleaseContents();
+ auto pipeline{CreateComputePipeline(ctx->pools, key, env)};
+ std::scoped_lock lock{state.mutex};
+ if (pipeline) {
+ compute_cache.emplace(key, std::move(pipeline));
+ }
+ ++state.built;
+ if (state.has_loaded) {
+ callback(VideoCore::LoadCallbackStage::Build, state.built, state.total);
+ }
+ });
++state.total;
}};
const auto load_graphics{[&](std::ifstream& file, std::vector<FileEnvironment> envs) {
GraphicsPipelineKey key;
file.read(reinterpret_cast<char*>(&key), sizeof(key));
- workers->QueueWork(
- [this, key, envs = std::move(envs), &state, &callback](Context* ctx) mutable {
- boost::container::static_vector<Shader::Environment*, 5> env_ptrs;
- for (auto& env : envs) {
- env_ptrs.push_back(&env);
- }
- ctx->pools.ReleaseContents();
- auto pipeline{CreateGraphicsPipeline(ctx->pools, key, MakeSpan(env_ptrs), false)};
- std::scoped_lock lock{state.mutex};
- if (pipeline) {
- graphics_cache.emplace(key, std::move(pipeline));
- }
- ++state.built;
- if (state.has_loaded) {
- callback(VideoCore::LoadCallbackStage::Build, state.built, state.total);
- }
- });
+ queue_work([this, key, envs = std::move(envs), &state, &callback](Context* ctx) mutable {
+ boost::container::static_vector<Shader::Environment*, 5> env_ptrs;
+ for (auto& env : envs) {
+ env_ptrs.push_back(&env);
+ }
+ ctx->pools.ReleaseContents();
+ auto pipeline{CreateGraphicsPipeline(ctx->pools, key, MakeSpan(env_ptrs), false)};
+ std::scoped_lock lock{state.mutex};
+ if (pipeline) {
+ graphics_cache.emplace(key, std::move(pipeline));
+ }
+ ++state.built;
+ if (state.has_loaded) {
+ callback(VideoCore::LoadCallbackStage::Build, state.built, state.total);
+ }
+ });
++state.total;
}};
LoadPipelines(stop_loading, shader_cache_filename, CACHE_VERSION, load_compute, load_graphics);
@@ -314,6 +325,9 @@ void ShaderCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading,
state.has_loaded = true;
lock.unlock();
+ if (strict_context_required) {
+ return;
+ }
workers->WaitForRequests(stop_loading);
if (!use_asynchronous_shaders) {
workers.reset();
diff --git a/src/video_core/renderer_opengl/gl_shader_cache.h b/src/video_core/renderer_opengl/gl_shader_cache.h
index 53ffea904..f82420592 100644
--- a/src/video_core/renderer_opengl/gl_shader_cache.h
+++ b/src/video_core/renderer_opengl/gl_shader_cache.h
@@ -69,6 +69,7 @@ private:
StateTracker& state_tracker;
VideoCore::ShaderNotify& shader_notify;
const bool use_asynchronous_shaders;
+ const bool strict_context_required;
GraphicsPipelineKey graphics_key{};
GraphicsPipeline* current_pipeline{};
diff --git a/src/video_core/renderer_opengl/renderer_opengl.cpp b/src/video_core/renderer_opengl/renderer_opengl.cpp
index 5b5e178ad..bc75680f0 100644
--- a/src/video_core/renderer_opengl/renderer_opengl.cpp
+++ b/src/video_core/renderer_opengl/renderer_opengl.cpp
@@ -140,8 +140,8 @@ RendererOpenGL::RendererOpenGL(Core::TelemetrySession& telemetry_session_,
Core::Memory::Memory& cpu_memory_, Tegra::GPU& gpu_,
std::unique_ptr<Core::Frontend::GraphicsContext> context_)
: RendererBase{emu_window_, std::move(context_)}, telemetry_session{telemetry_session_},
- emu_window{emu_window_}, cpu_memory{cpu_memory_}, gpu{gpu_}, state_tracker{},
- program_manager{device},
+ emu_window{emu_window_}, cpu_memory{cpu_memory_}, gpu{gpu_}, device{emu_window_},
+ state_tracker{}, program_manager{device},
rasterizer(emu_window, gpu, cpu_memory, device, screen_info, program_manager, state_tracker) {
if (Settings::values.renderer_debug && GLAD_GL_KHR_debug) {
glEnable(GL_DEBUG_OUTPUT);
diff --git a/src/video_core/renderer_vulkan/renderer_vulkan.cpp b/src/video_core/renderer_vulkan/renderer_vulkan.cpp
index 18be54729..f502a7d09 100644
--- a/src/video_core/renderer_vulkan/renderer_vulkan.cpp
+++ b/src/video_core/renderer_vulkan/renderer_vulkan.cpp
@@ -139,23 +139,25 @@ void RendererVulkan::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
RenderScreenshot(*framebuffer, use_accelerated);
bool has_been_recreated = false;
- const auto recreate_swapchain = [&] {
+ const auto recreate_swapchain = [&](u32 width, u32 height) {
if (!has_been_recreated) {
has_been_recreated = true;
scheduler.Finish();
}
- const Layout::FramebufferLayout layout = render_window.GetFramebufferLayout();
- swapchain.Create(layout.width, layout.height, is_srgb);
+ swapchain.Create(width, height, is_srgb);
};
- if (swapchain.NeedsRecreation(is_srgb)) {
- recreate_swapchain();
+
+ const Layout::FramebufferLayout layout = render_window.GetFramebufferLayout();
+ if (swapchain.NeedsRecreation(is_srgb) || swapchain.GetWidth() != layout.width ||
+ swapchain.GetHeight() != layout.height) {
+ recreate_swapchain(layout.width, layout.height);
}
bool is_outdated;
do {
swapchain.AcquireNextImage();
is_outdated = swapchain.IsOutDated();
if (is_outdated) {
- recreate_swapchain();
+ recreate_swapchain(layout.width, layout.height);
}
} while (is_outdated);
if (has_been_recreated) {
diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp
index 558b8db56..84d36fea6 100644
--- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp
@@ -285,6 +285,9 @@ void BufferCacheRuntime::BindQuadArrayIndexBuffer(u32 first, u32 count) {
void BufferCacheRuntime::BindVertexBuffer(u32 index, VkBuffer buffer, u32 offset, u32 size,
u32 stride) {
+ if (index >= device.GetMaxVertexInputBindings()) {
+ return;
+ }
if (device.IsExtExtendedDynamicStateSupported()) {
scheduler.Record([index, buffer, offset, size, stride](vk::CommandBuffer cmdbuf) {
const VkDeviceSize vk_offset = buffer != VK_NULL_HANDLE ? offset : 0;
diff --git a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
index 006128638..4b10fe7bc 100644
--- a/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
+++ b/src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
@@ -529,7 +529,9 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) {
static_vector<VkVertexInputBindingDivisorDescriptionEXT, 32> vertex_binding_divisors;
static_vector<VkVertexInputAttributeDescription, 32> vertex_attributes;
if (key.state.dynamic_vertex_input) {
- for (size_t index = 0; index < key.state.attributes.size(); ++index) {
+ const size_t num_vertex_arrays = std::min(
+ key.state.attributes.size(), static_cast<size_t>(device.GetMaxVertexInputBindings()));
+ for (size_t index = 0; index < num_vertex_arrays; ++index) {
const u32 type = key.state.DynamicAttributeType(index);
if (!stage_infos[0].loads.Generic(index) || type == 0) {
continue;
@@ -551,7 +553,9 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) {
});
}
} else {
- for (size_t index = 0; index < Maxwell::NumVertexArrays; ++index) {
+ const size_t num_vertex_arrays = std::min(
+ Maxwell::NumVertexArrays, static_cast<size_t>(device.GetMaxVertexInputBindings()));
+ for (size_t index = 0; index < num_vertex_arrays; ++index) {
const bool instanced = key.state.binding_divisors[index] != 0;
const auto rate =
instanced ? VK_VERTEX_INPUT_RATE_INSTANCE : VK_VERTEX_INPUT_RATE_VERTEX;
@@ -580,6 +584,8 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) {
});
}
}
+ ASSERT(vertex_attributes.size() <= device.GetMaxVertexInputAttributes());
+
VkPipelineVertexInputStateCreateInfo vertex_input_ci{
.sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
.pNext = nullptr,
diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
index 81f5f3e11..86fdde014 100644
--- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
@@ -341,6 +341,15 @@ PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, const Device& device
.support_snorm_render_buffer = true,
.support_viewport_index_layer = device.IsExtShaderViewportIndexLayerSupported(),
};
+
+ if (device.GetMaxVertexInputAttributes() < Maxwell::NumVertexAttributes) {
+ LOG_WARNING(Render_Vulkan, "maxVertexInputAttributes is too low: {} < {}",
+ device.GetMaxVertexInputAttributes(), Maxwell::NumVertexAttributes);
+ }
+ if (device.GetMaxVertexInputBindings() < Maxwell::NumVertexArrays) {
+ LOG_WARNING(Render_Vulkan, "maxVertexInputBindings is too low: {} < {}",
+ device.GetMaxVertexInputBindings(), Maxwell::NumVertexArrays);
+ }
}
PipelineCache::~PipelineCache() = default;
diff --git a/src/video_core/renderer_vulkan/vk_swapchain.cpp b/src/video_core/renderer_vulkan/vk_swapchain.cpp
index d7be417f5..b6810eef9 100644
--- a/src/video_core/renderer_vulkan/vk_swapchain.cpp
+++ b/src/video_core/renderer_vulkan/vk_swapchain.cpp
@@ -67,17 +67,19 @@ VkExtent2D ChooseSwapExtent(const VkSurfaceCapabilitiesKHR& capabilities, u32 wi
} // Anonymous namespace
-Swapchain::Swapchain(VkSurfaceKHR surface_, const Device& device_, Scheduler& scheduler_, u32 width,
- u32 height, bool srgb)
+Swapchain::Swapchain(VkSurfaceKHR surface_, const Device& device_, Scheduler& scheduler_,
+ u32 width_, u32 height_, bool srgb)
: surface{surface_}, device{device_}, scheduler{scheduler_} {
- Create(width, height, srgb);
+ Create(width_, height_, srgb);
}
Swapchain::~Swapchain() = default;
-void Swapchain::Create(u32 width, u32 height, bool srgb) {
+void Swapchain::Create(u32 width_, u32 height_, bool srgb) {
is_outdated = false;
is_suboptimal = false;
+ width = width_;
+ height = height_;
const auto physical_device = device.GetPhysical();
const auto capabilities{physical_device.GetSurfaceCapabilitiesKHR(surface)};
@@ -88,7 +90,7 @@ void Swapchain::Create(u32 width, u32 height, bool srgb) {
device.GetLogical().WaitIdle();
Destroy();
- CreateSwapchain(capabilities, width, height, srgb);
+ CreateSwapchain(capabilities, srgb);
CreateSemaphores();
CreateImageViews();
@@ -148,8 +150,7 @@ void Swapchain::Present(VkSemaphore render_semaphore) {
}
}
-void Swapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, u32 width, u32 height,
- bool srgb) {
+void Swapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, bool srgb) {
const auto physical_device{device.GetPhysical()};
const auto formats{physical_device.GetSurfaceFormatsKHR(surface)};
const auto present_modes{physical_device.GetSurfacePresentModesKHR(surface)};
diff --git a/src/video_core/renderer_vulkan/vk_swapchain.h b/src/video_core/renderer_vulkan/vk_swapchain.h
index 111b3902d..caf1ff32b 100644
--- a/src/video_core/renderer_vulkan/vk_swapchain.h
+++ b/src/video_core/renderer_vulkan/vk_swapchain.h
@@ -80,9 +80,16 @@ public:
return *present_semaphores[frame_index];
}
+ u32 GetWidth() const {
+ return width;
+ }
+
+ u32 GetHeight() const {
+ return height;
+ }
+
private:
- void CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, u32 width, u32 height,
- bool srgb);
+ void CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, bool srgb);
void CreateSemaphores();
void CreateImageViews();
@@ -105,6 +112,9 @@ private:
std::vector<u64> resource_ticks;
std::vector<vk::Semaphore> present_semaphores;
+ u32 width;
+ u32 height;
+
u32 image_index{};
u32 frame_index{};
diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp
index 6a2ad4b1d..f45030311 100644
--- a/src/video_core/vulkan_common/vulkan_device.cpp
+++ b/src/video_core/vulkan_common/vulkan_device.cpp
@@ -421,7 +421,7 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
VkPhysicalDevice8BitStorageFeatures bit8_storage{
.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_8BIT_STORAGE_FEATURES,
.pNext = nullptr,
- .storageBuffer8BitAccess = false,
+ .storageBuffer8BitAccess = true,
.uniformAndStorageBuffer8BitAccess = true,
.storagePushConstant8 = false,
};
@@ -1044,6 +1044,7 @@ void Device::CheckSuitability(bool requires_swapchain) const {
std::make_pair(bit16_storage.storageBuffer16BitAccess, "storageBuffer16BitAccess"),
std::make_pair(bit16_storage.uniformAndStorageBuffer16BitAccess,
"uniformAndStorageBuffer16BitAccess"),
+ std::make_pair(bit8_storage.storageBuffer8BitAccess, "storageBuffer8BitAccess"),
std::make_pair(bit8_storage.uniformAndStorageBuffer8BitAccess,
"uniformAndStorageBuffer8BitAccess"),
std::make_pair(host_query_reset.hostQueryReset, "hostQueryReset"),
@@ -1380,6 +1381,10 @@ void Device::SetupFeatures() {
is_shader_storage_image_multisample = features.shaderStorageImageMultisample;
is_blit_depth_stencil_supported = TestDepthStencilBlits();
is_optimal_astc_supported = IsOptimalAstcSupported(features);
+
+ const VkPhysicalDeviceLimits& limits{properties.limits};
+ max_vertex_input_attributes = limits.maxVertexInputAttributes;
+ max_vertex_input_bindings = limits.maxVertexInputBindings;
}
void Device::SetupProperties() {
diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h
index db802437c..391b7604c 100644
--- a/src/video_core/vulkan_common/vulkan_device.h
+++ b/src/video_core/vulkan_common/vulkan_device.h
@@ -368,6 +368,14 @@ public:
return must_emulate_bgr565;
}
+ u32 GetMaxVertexInputAttributes() const {
+ return max_vertex_input_attributes;
+ }
+
+ u32 GetMaxVertexInputBindings() const {
+ return max_vertex_input_bindings;
+ }
+
private:
/// Checks if the physical device is suitable.
void CheckSuitability(bool requires_swapchain) const;
@@ -467,6 +475,8 @@ private:
bool supports_d24_depth{}; ///< Supports D24 depth buffers.
bool cant_blit_msaa{}; ///< Does not support MSAA<->MSAA blitting.
bool must_emulate_bgr565{}; ///< Emulates BGR565 by swizzling RGB565 format.
+ u32 max_vertex_input_attributes{}; ///< Max vertex input attributes in pipeline
+ u32 max_vertex_input_bindings{}; ///< Max vertex input buffers in pipeline
// Telemetry parameters
std::string vendor_name; ///< Device's driver name.
generated by cgit v1.2.3 (git 2.25.1) at 2024-07-19 16:11:42 +0200