summaryrefslogtreecommitdiffstats
path: root/src/video_core
diff options
context:
space:
mode:
Diffstat (limited to 'src/video_core')
-rw-r--r--src/video_core/engines/maxwell_dma.cpp1
-rw-r--r--src/video_core/renderer_opengl/gl_device.cpp43
-rw-r--r--src/video_core/renderer_opengl/gl_shader_cache.cpp10
-rw-r--r--src/video_core/renderer_vulkan/vk_pipeline_cache.cpp12
-rw-r--r--src/video_core/shader_cache.cpp5
-rw-r--r--src/video_core/shader_cache.h2
-rw-r--r--src/video_core/shader_environment.cpp31
-rw-r--r--src/video_core/shader_environment.h6
-rw-r--r--src/video_core/vulkan_common/vulkan_device.cpp181
-rw-r--r--src/video_core/vulkan_common/vulkan_device.h13
10 files changed, 201 insertions, 103 deletions
diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp
index cd8e24b0b..da8eab7ee 100644
--- a/src/video_core/engines/maxwell_dma.cpp
+++ b/src/video_core/engines/maxwell_dma.cpp
@@ -5,6 +5,7 @@
#include "common/assert.h"
#include "common/logging/log.h"
#include "common/microprofile.h"
+#include "common/polyfill_ranges.h"
#include "common/settings.h"
#include "core/core.h"
#include "core/memory.h"
diff --git a/src/video_core/renderer_opengl/gl_device.cpp b/src/video_core/renderer_opengl/gl_device.cpp
index ee140c9c2..94258ccd0 100644
--- a/src/video_core/renderer_opengl/gl_device.cpp
+++ b/src/video_core/renderer_opengl/gl_device.cpp
@@ -106,6 +106,43 @@ bool IsASTCSupported() {
return true;
}
+static bool HasSlowSoftwareAstc(std::string_view vendor_name, std::string_view renderer) {
+// ifdef for Unix reduces string comparisons for non-Windows drivers, and Intel
+#ifdef YUZU_UNIX
+ // Sorted vaguely by how likely a vendor is to appear
+ if (vendor_name == "AMD") {
+ // RadeonSI
+ return true;
+ }
+ if (vendor_name == "Intel") {
+ // Must be inside YUZU_UNIX ifdef as the Windows driver uses the same vendor string
+ // iris, crocus
+ const bool is_intel_dg = (renderer.find("DG") != std::string_view::npos);
+ return is_intel_dg;
+ }
+ if (vendor_name == "nouveau") {
+ return true;
+ }
+ if (vendor_name == "X.Org") {
+ // R600
+ return true;
+ }
+#endif
+ if (vendor_name == "Collabora Ltd") {
+ // Zink
+ return true;
+ }
+ if (vendor_name == "Microsoft Corporation") {
+ // d3d12
+ return true;
+ }
+ if (vendor_name == "Mesa/X.org") {
+ // llvmpipe, softpipe, virgl
+ return true;
+ }
+ return false;
+}
+
[[nodiscard]] bool IsDebugToolAttached(std::span<const std::string_view> extensions) {
const bool nsight = std::getenv("NVTX_INJECTION64_PATH") || std::getenv("NSIGHT_LAUNCHED");
return nsight || HasExtension(extensions, "GL_EXT_debug_tool") ||
@@ -120,12 +157,16 @@ Device::Device(Core::Frontend::EmuWindow& emu_window) {
}
vendor_name = reinterpret_cast<const char*>(glGetString(GL_VENDOR));
const std::string_view version = reinterpret_cast<const char*>(glGetString(GL_VERSION));
+ const std::string_view renderer = reinterpret_cast<const char*>(glGetString(GL_RENDERER));
const std::vector extensions = GetExtensions();
const bool is_nvidia = vendor_name == "NVIDIA Corporation";
const bool is_amd = vendor_name == "ATI Technologies Inc.";
const bool is_intel = vendor_name == "Intel";
+ const bool has_slow_software_astc =
+ !is_nvidia && !is_amd && HasSlowSoftwareAstc(vendor_name, renderer);
+
#ifdef __unix__
constexpr bool is_linux = true;
#else
@@ -152,7 +193,7 @@ Device::Device(Core::Frontend::EmuWindow& emu_window) {
has_vertex_viewport_layer = GLAD_GL_ARB_shader_viewport_layer_array;
has_image_load_formatted = HasExtension(extensions, "GL_EXT_shader_image_load_formatted");
has_texture_shadow_lod = HasExtension(extensions, "GL_EXT_texture_shadow_lod");
- has_astc = IsASTCSupported();
+ has_astc = !has_slow_software_astc && IsASTCSupported();
has_variable_aoffi = TestVariableAoffi();
has_component_indexing_bug = is_amd;
has_precise_bug = TestPreciseBug();
diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp
index 618cb6354..2888e0238 100644
--- a/src/video_core/renderer_opengl/gl_shader_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp
@@ -445,7 +445,8 @@ std::unique_ptr<GraphicsPipeline> ShaderCache::CreateGraphicsPipeline(
ShaderContext::ShaderPools& pools, const GraphicsPipelineKey& key,
std::span<Shader::Environment* const> envs, bool use_shader_workers,
bool force_context_flush) try {
- LOG_INFO(Render_OpenGL, "0x{:016x}", key.Hash());
+ auto hash = key.Hash();
+ LOG_INFO(Render_OpenGL, "0x{:016x}", hash);
size_t env_index{};
u32 total_storage_buffers{};
std::array<Shader::IR::Program, Maxwell::MaxShaderProgram> programs;
@@ -474,7 +475,7 @@ std::unique_ptr<GraphicsPipeline> ShaderCache::CreateGraphicsPipeline(
Shader::Maxwell::Flow::CFG cfg(env, pools.flow_block, cfg_offset, index == 0);
if (Settings::values.dump_shaders) {
- env.Dump(key.unique_hashes[index]);
+ env.Dump(hash, key.unique_hashes[index]);
}
if (!uses_vertex_a || index != 1) {
@@ -566,12 +567,13 @@ std::unique_ptr<ComputePipeline> ShaderCache::CreateComputePipeline(
std::unique_ptr<ComputePipeline> ShaderCache::CreateComputePipeline(
ShaderContext::ShaderPools& pools, const ComputePipelineKey& key, Shader::Environment& env,
bool force_context_flush) try {
- LOG_INFO(Render_OpenGL, "0x{:016x}", key.Hash());
+ auto hash = key.Hash();
+ LOG_INFO(Render_OpenGL, "0x{:016x}", hash);
Shader::Maxwell::Flow::CFG cfg{env, pools.flow_block, env.StartAddress()};
if (Settings::values.dump_shaders) {
- env.Dump(key.Hash());
+ env.Dump(hash, key.unique_hash);
}
auto program{TranslateProgram(pools.inst, pools.block, env, cfg, host_info)};
diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
index 4f84d8497..c1314ca99 100644
--- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
@@ -584,7 +584,8 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline(
ShaderPools& pools, const GraphicsPipelineCacheKey& key,
std::span<Shader::Environment* const> envs, PipelineStatistics* statistics,
bool build_in_parallel) try {
- LOG_INFO(Render_Vulkan, "0x{:016x}", key.Hash());
+ auto hash = key.Hash();
+ LOG_INFO(Render_Vulkan, "0x{: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};
@@ -611,7 +612,7 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline(
const u32 cfg_offset{static_cast<u32>(env.StartAddress() + sizeof(Shader::ProgramHeader))};
Shader::Maxwell::Flow::CFG cfg(env, pools.flow_block, cfg_offset, index == 0);
if (Settings::values.dump_shaders) {
- env.Dump(key.unique_hashes[index]);
+ env.Dump(hash, key.unique_hashes[index]);
}
if (!uses_vertex_a || index != 1) {
// Normal path
@@ -712,18 +713,19 @@ std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
ShaderPools& pools, const ComputePipelineCacheKey& key, Shader::Environment& env,
PipelineStatistics* statistics, bool build_in_parallel) try {
+ auto hash = key.Hash();
if (device.HasBrokenCompute()) {
- LOG_ERROR(Render_Vulkan, "Skipping 0x{:016x}", key.Hash());
+ LOG_ERROR(Render_Vulkan, "Skipping 0x{:016x}", hash);
return nullptr;
}
- LOG_INFO(Render_Vulkan, "0x{:016x}", key.Hash());
+ LOG_INFO(Render_Vulkan, "0x{:016x}", hash);
Shader::Maxwell::Flow::CFG cfg{env, pools.flow_block, env.StartAddress()};
// Dump it before error.
if (Settings::values.dump_shaders) {
- env.Dump(key.Hash());
+ env.Dump(hash, key.unique_hash);
}
auto program{TranslateProgram(pools.inst, pools.block, env, cfg, host_info)};
diff --git a/src/video_core/shader_cache.cpp b/src/video_core/shader_cache.cpp
index 01701201d..e81cd031b 100644
--- a/src/video_core/shader_cache.cpp
+++ b/src/video_core/shader_cache.cpp
@@ -51,6 +51,11 @@ bool ShaderCache::RefreshStages(std::array<u64, 6>& unique_hashes) {
}
const auto& shader_config{maxwell3d->regs.pipelines[index]};
const auto program{static_cast<Tegra::Engines::Maxwell3D::Regs::ShaderType>(index)};
+ if (program == Tegra::Engines::Maxwell3D::Regs::ShaderType::Pixel &&
+ !maxwell3d->regs.rasterize_enable) {
+ unique_hashes[index] = 0;
+ continue;
+ }
const GPUVAddr shader_addr{base_addr + shader_config.offset};
const std::optional<VAddr> cpu_shader_addr{gpu_memory->GpuToCpuAddress(shader_addr)};
if (!cpu_shader_addr) {
diff --git a/src/video_core/shader_cache.h b/src/video_core/shader_cache.h
index de8e08002..a76896620 100644
--- a/src/video_core/shader_cache.h
+++ b/src/video_core/shader_cache.h
@@ -70,7 +70,7 @@ public:
protected:
struct GraphicsEnvironments {
std::array<GraphicsEnvironment, NUM_PROGRAMS> envs;
- std::array<Shader::Environment*, NUM_PROGRAMS> env_ptrs;
+ std::array<Shader::Environment*, NUM_PROGRAMS> env_ptrs{};
std::span<Shader::Environment* const> Span() const noexcept {
return std::span(env_ptrs.begin(), std::ranges::find(env_ptrs, nullptr));
diff --git a/src/video_core/shader_environment.cpp b/src/video_core/shader_environment.cpp
index c7cb56243..4edbe5700 100644
--- a/src/video_core/shader_environment.cpp
+++ b/src/video_core/shader_environment.cpp
@@ -102,7 +102,8 @@ static std::string_view StageToPrefix(Shader::Stage stage) {
}
}
-static void DumpImpl(u64 hash, const u64* code, u32 read_highest, u32 read_lowest,
+static void DumpImpl(u64 pipeline_hash, u64 shader_hash, std::span<const u64> code,
+ [[maybe_unused]] u32 read_highest, [[maybe_unused]] u32 read_lowest,
u32 initial_offset, Shader::Stage stage) {
const auto shader_dir{Common::FS::GetYuzuPath(Common::FS::YuzuPath::DumpDir)};
const auto base_dir{shader_dir / "shaders"};
@@ -111,13 +112,18 @@ static void DumpImpl(u64 hash, const u64* code, u32 read_highest, u32 read_lowes
return;
}
const auto prefix = StageToPrefix(stage);
- const auto name{base_dir / fmt::format("{}{:016x}.ash", prefix, hash)};
- const size_t real_size = read_highest - read_lowest + initial_offset;
- const size_t padding_needed = ((32 - (real_size % 32)) % 32);
+ const auto name{base_dir /
+ fmt::format("{:016x}_{}_{:016x}.ash", pipeline_hash, prefix, shader_hash)};
std::fstream shader_file(name, std::ios::out | std::ios::binary);
+ ASSERT(initial_offset % sizeof(u64) == 0);
const size_t jump_index = initial_offset / sizeof(u64);
- shader_file.write(reinterpret_cast<const char*>(code + jump_index), real_size);
- for (size_t i = 0; i < padding_needed; i++) {
+ const size_t code_size = code.size_bytes() - initial_offset;
+ shader_file.write(reinterpret_cast<const char*>(&code[jump_index]), code_size);
+
+ // + 1 instruction, due to the fact that we skip the final self branch instruction in the code,
+ // but we need to consider it for padding, otherwise nvdisasm rages.
+ const size_t padding_needed = (32 - ((code_size + INST_SIZE) % 32)) % 32;
+ for (size_t i = 0; i < INST_SIZE + padding_needed; i++) {
shader_file.put(0);
}
}
@@ -197,8 +203,8 @@ u64 GenericEnvironment::CalculateHash() const {
return Common::CityHash64(data.get(), size);
}
-void GenericEnvironment::Dump(u64 hash) {
- DumpImpl(hash, code.data(), read_highest, read_lowest, initial_offset, stage);
+void GenericEnvironment::Dump(u64 pipeline_hash, u64 shader_hash) {
+ DumpImpl(pipeline_hash, shader_hash, code, read_highest, read_lowest, initial_offset, stage);
}
void GenericEnvironment::Serialize(std::ofstream& file) const {
@@ -282,6 +288,7 @@ std::optional<u64> GenericEnvironment::TryFindSize() {
Tegra::Texture::TICEntry GenericEnvironment::ReadTextureInfo(GPUVAddr tic_addr, u32 tic_limit,
bool via_header_index, u32 raw) {
const auto handle{Tegra::Texture::TexturePair(raw, via_header_index)};
+ ASSERT(handle.first <= tic_limit);
const GPUVAddr descriptor_addr{tic_addr + handle.first * sizeof(Tegra::Texture::TICEntry)};
Tegra::Texture::TICEntry entry;
gpu_memory->ReadBlock(descriptor_addr, &entry, sizeof(entry));
@@ -465,8 +472,8 @@ void FileEnvironment::Deserialize(std::ifstream& file) {
.read(reinterpret_cast<char*>(&read_highest), sizeof(read_highest))
.read(reinterpret_cast<char*>(&viewport_transform_state), sizeof(viewport_transform_state))
.read(reinterpret_cast<char*>(&stage), sizeof(stage));
- code = std::make_unique<u64[]>(Common::DivCeil(code_size, sizeof(u64)));
- file.read(reinterpret_cast<char*>(code.get()), code_size);
+ code.resize(Common::DivCeil(code_size, sizeof(u64)));
+ file.read(reinterpret_cast<char*>(code.data()), code_size);
for (size_t i = 0; i < num_texture_types; ++i) {
u32 key;
Shader::TextureType type;
@@ -509,8 +516,8 @@ void FileEnvironment::Deserialize(std::ifstream& file) {
is_propietary_driver = texture_bound == 2;
}
-void FileEnvironment::Dump(u64 hash) {
- DumpImpl(hash, code.get(), read_highest, read_lowest, initial_offset, stage);
+void FileEnvironment::Dump(u64 pipeline_hash, u64 shader_hash) {
+ DumpImpl(pipeline_hash, shader_hash, code, read_highest, read_lowest, initial_offset, stage);
}
u64 FileEnvironment::ReadInstruction(u32 address) {
diff --git a/src/video_core/shader_environment.h b/src/video_core/shader_environment.h
index a0f61cbda..b90f3d44e 100644
--- a/src/video_core/shader_environment.h
+++ b/src/video_core/shader_environment.h
@@ -58,7 +58,7 @@ public:
[[nodiscard]] u64 CalculateHash() const;
- void Dump(u64 hash) override;
+ void Dump(u64 pipeline_hash, u64 shader_hash) override;
void Serialize(std::ofstream& file) const;
@@ -188,10 +188,10 @@ public:
return cbuf_replacements.size() != 0;
}
- void Dump(u64 hash) override;
+ void Dump(u64 pipeline_hash, u64 shader_hash) override;
private:
- std::unique_ptr<u64[]> code;
+ std::vector<u64> code;
std::unordered_map<u32, Shader::TextureType> texture_types;
std::unordered_map<u32, Shader::TexturePixelFormat> texture_pixel_formats;
std::unordered_map<u64, u32> cbuf_values;
diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp
index bd67e27ed..adde96aa5 100644
--- a/src/video_core/vulkan_common/vulkan_device.cpp
+++ b/src/video_core/vulkan_common/vulkan_device.cpp
@@ -326,6 +326,43 @@ std::vector<const char*> ExtensionListForVulkan(
} // Anonymous namespace
+void Device::RemoveExtension(bool& extension, const std::string& extension_name) {
+ extension = false;
+ loaded_extensions.erase(extension_name);
+}
+
+void Device::RemoveExtensionIfUnsuitable(bool is_suitable, const std::string& extension_name) {
+ if (loaded_extensions.contains(extension_name) && !is_suitable) {
+ LOG_WARNING(Render_Vulkan, "Removing unsuitable extension {}", extension_name);
+ this->RemoveExtension(is_suitable, extension_name);
+ }
+}
+
+template <typename Feature>
+void Device::RemoveExtensionFeature(bool& extension, Feature& feature,
+ const std::string& extension_name) {
+ // Unload extension.
+ this->RemoveExtension(extension, extension_name);
+
+ // Save sType and pNext for chain.
+ VkStructureType sType = feature.sType;
+ void* pNext = feature.pNext;
+
+ // Clear feature struct and restore chain.
+ feature = {};
+ feature.sType = sType;
+ feature.pNext = pNext;
+}
+
+template <typename Feature>
+void Device::RemoveExtensionFeatureIfUnsuitable(bool is_suitable, Feature& feature,
+ const std::string& extension_name) {
+ if (loaded_extensions.contains(extension_name) && !is_suitable) {
+ LOG_WARNING(Render_Vulkan, "Removing features for unsuitable extension {}", extension_name);
+ this->RemoveExtensionFeature(is_suitable, feature, extension_name);
+ }
+}
+
Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR surface,
const vk::InstanceDispatch& dld_)
: instance{instance_}, dld{dld_}, physical{physical_},
@@ -397,21 +434,20 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
if (is_qualcomm || is_turnip) {
LOG_WARNING(Render_Vulkan,
"Qualcomm and Turnip drivers have broken VK_EXT_custom_border_color");
- extensions.custom_border_color = false;
- loaded_extensions.erase(VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME);
+ RemoveExtensionFeature(extensions.custom_border_color, features.custom_border_color,
+ VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME);
}
if (is_qualcomm) {
must_emulate_scaled_formats = true;
LOG_WARNING(Render_Vulkan, "Qualcomm drivers have broken VK_EXT_extended_dynamic_state");
- extensions.extended_dynamic_state = false;
- loaded_extensions.erase(VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME);
+ RemoveExtensionFeature(extensions.extended_dynamic_state, features.extended_dynamic_state,
+ VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME);
LOG_WARNING(Render_Vulkan,
"Qualcomm drivers have a slow VK_KHR_push_descriptor implementation");
- extensions.push_descriptor = false;
- loaded_extensions.erase(VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME);
+ RemoveExtension(extensions.push_descriptor, VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME);
#if defined(ANDROID) && defined(ARCHITECTURE_arm64)
// Patch the driver to enable BCn textures.
@@ -440,15 +476,12 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
must_emulate_scaled_formats = true;
LOG_WARNING(Render_Vulkan, "ARM drivers have broken VK_EXT_extended_dynamic_state");
- extensions.extended_dynamic_state = false;
- loaded_extensions.erase(VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME);
+ RemoveExtensionFeature(extensions.extended_dynamic_state, features.extended_dynamic_state,
+ VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME);
LOG_WARNING(Render_Vulkan, "ARM drivers have broken VK_EXT_extended_dynamic_state2");
- features.extended_dynamic_state2.extendedDynamicState2 = false;
- features.extended_dynamic_state2.extendedDynamicState2LogicOp = false;
- features.extended_dynamic_state2.extendedDynamicState2PatchControlPoints = false;
- extensions.extended_dynamic_state2 = false;
- loaded_extensions.erase(VK_EXT_EXTENDED_DYNAMIC_STATE_2_EXTENSION_NAME);
+ RemoveExtensionFeature(extensions.extended_dynamic_state2, features.extended_dynamic_state2,
+ VK_EXT_EXTENDED_DYNAMIC_STATE_2_EXTENSION_NAME);
}
if (is_nvidia) {
@@ -464,8 +497,7 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
case NvidiaArchitecture::VoltaOrOlder:
if (nv_major_version < 527) {
LOG_WARNING(Render_Vulkan, "Volta and older have broken VK_KHR_push_descriptor");
- extensions.push_descriptor = false;
- loaded_extensions.erase(VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME);
+ RemoveExtension(extensions.push_descriptor, VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME);
}
break;
}
@@ -480,8 +512,9 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
if (version < VK_MAKE_API_VERSION(0, 21, 2, 0)) {
LOG_WARNING(Render_Vulkan,
"RADV versions older than 21.2 have broken VK_EXT_extended_dynamic_state");
- extensions.extended_dynamic_state = false;
- loaded_extensions.erase(VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME);
+ RemoveExtensionFeature(extensions.extended_dynamic_state,
+ features.extended_dynamic_state,
+ VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME);
}
}
if (extensions.extended_dynamic_state2 && is_radv) {
@@ -490,11 +523,9 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
LOG_WARNING(
Render_Vulkan,
"RADV versions older than 22.3.1 have broken VK_EXT_extended_dynamic_state2");
- features.extended_dynamic_state2.extendedDynamicState2 = false;
- features.extended_dynamic_state2.extendedDynamicState2LogicOp = false;
- features.extended_dynamic_state2.extendedDynamicState2PatchControlPoints = false;
- extensions.extended_dynamic_state2 = false;
- loaded_extensions.erase(VK_EXT_EXTENDED_DYNAMIC_STATE_2_EXTENSION_NAME);
+ RemoveExtensionFeature(extensions.extended_dynamic_state2,
+ features.extended_dynamic_state2,
+ VK_EXT_EXTENDED_DYNAMIC_STATE_2_EXTENSION_NAME);
}
}
if (extensions.extended_dynamic_state2 && is_qualcomm) {
@@ -504,11 +535,9 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
// Qualcomm Adreno 7xx drivers do not properly support extended_dynamic_state2.
LOG_WARNING(Render_Vulkan,
"Qualcomm Adreno 7xx drivers have broken VK_EXT_extended_dynamic_state2");
- features.extended_dynamic_state2.extendedDynamicState2 = false;
- features.extended_dynamic_state2.extendedDynamicState2LogicOp = false;
- features.extended_dynamic_state2.extendedDynamicState2PatchControlPoints = false;
- extensions.extended_dynamic_state2 = false;
- loaded_extensions.erase(VK_EXT_EXTENDED_DYNAMIC_STATE_2_EXTENSION_NAME);
+ RemoveExtensionFeature(extensions.extended_dynamic_state2,
+ features.extended_dynamic_state2,
+ VK_EXT_EXTENDED_DYNAMIC_STATE_2_EXTENSION_NAME);
}
}
if (extensions.extended_dynamic_state3 && is_radv) {
@@ -540,9 +569,9 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
if (is_rdna2) {
LOG_WARNING(Render_Vulkan,
"RADV has broken VK_EXT_vertex_input_dynamic_state on RDNA2 hardware");
- features.vertex_input_dynamic_state.vertexInputDynamicState = false;
- extensions.vertex_input_dynamic_state = false;
- loaded_extensions.erase(VK_EXT_VERTEX_INPUT_DYNAMIC_STATE_EXTENSION_NAME);
+ RemoveExtensionFeature(extensions.vertex_input_dynamic_state,
+ features.vertex_input_dynamic_state,
+ VK_EXT_VERTEX_INPUT_DYNAMIC_STATE_EXTENSION_NAME);
}
}
if (extensions.vertex_input_dynamic_state && is_qualcomm) {
@@ -553,9 +582,9 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
LOG_WARNING(
Render_Vulkan,
"Qualcomm Adreno 7xx drivers have broken VK_EXT_vertex_input_dynamic_state");
- features.vertex_input_dynamic_state.vertexInputDynamicState = false;
- extensions.vertex_input_dynamic_state = false;
- loaded_extensions.erase(VK_EXT_VERTEX_INPUT_DYNAMIC_STATE_EXTENSION_NAME);
+ RemoveExtensionFeature(extensions.vertex_input_dynamic_state,
+ features.vertex_input_dynamic_state,
+ VK_EXT_VERTEX_INPUT_DYNAMIC_STATE_EXTENSION_NAME);
}
}
@@ -575,8 +604,8 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
if (!features.shader_float16_int8.shaderFloat16) {
LOG_WARNING(Render_Vulkan,
"AMD GCN4 and earlier have broken VK_EXT_sampler_filter_minmax");
- extensions.sampler_filter_minmax = false;
- loaded_extensions.erase(VK_EXT_SAMPLER_FILTER_MINMAX_EXTENSION_NAME);
+ RemoveExtension(extensions.sampler_filter_minmax,
+ VK_EXT_SAMPLER_FILTER_MINMAX_EXTENSION_NAME);
}
}
@@ -584,8 +613,9 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
const u32 version = (properties.properties.driverVersion << 3) >> 3;
if (version < VK_MAKE_API_VERSION(27, 20, 100, 0)) {
LOG_WARNING(Render_Vulkan, "Intel has broken VK_EXT_vertex_input_dynamic_state");
- extensions.vertex_input_dynamic_state = false;
- loaded_extensions.erase(VK_EXT_VERTEX_INPUT_DYNAMIC_STATE_EXTENSION_NAME);
+ RemoveExtensionFeature(extensions.vertex_input_dynamic_state,
+ features.vertex_input_dynamic_state,
+ VK_EXT_VERTEX_INPUT_DYNAMIC_STATE_EXTENSION_NAME);
}
}
if (features.shader_float16_int8.shaderFloat16 && is_intel_windows) {
@@ -612,8 +642,7 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
// mesa/mesa/-/commit/ff91c5ca42bc80aa411cb3fd8f550aa6fdd16bdc
LOG_WARNING(Render_Vulkan,
"ANV drivers 22.3.0 to 23.1.0 have broken VK_KHR_push_descriptor");
- extensions.push_descriptor = false;
- loaded_extensions.erase(VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME);
+ RemoveExtension(extensions.push_descriptor, VK_KHR_PUSH_DESCRIPTOR_EXTENSION_NAME);
}
}
if (is_mvk) {
@@ -963,7 +992,7 @@ bool Device::GetSuitability(bool requires_swapchain) {
VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PUSH_DESCRIPTOR_PROPERTIES_KHR;
SetNext(next, properties.push_descriptor);
}
- if (extensions.subgroup_size_control) {
+ if (extensions.subgroup_size_control || features.subgroup_size_control.subgroupSizeControl) {
properties.subgroup_size_control.sType =
VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_SIZE_CONTROL_PROPERTIES;
SetNext(next, properties.subgroup_size_control);
@@ -1007,34 +1036,29 @@ bool Device::GetSuitability(bool requires_swapchain) {
return suitable;
}
-void Device::RemoveExtensionIfUnsuitable(bool is_suitable, const std::string& extension_name) {
- if (loaded_extensions.contains(extension_name) && !is_suitable) {
- LOG_WARNING(Render_Vulkan, "Removing unsuitable extension {}", extension_name);
- loaded_extensions.erase(extension_name);
- }
-}
-
void Device::RemoveUnsuitableExtensions() {
// VK_EXT_custom_border_color
extensions.custom_border_color = features.custom_border_color.customBorderColors &&
features.custom_border_color.customBorderColorWithoutFormat;
- RemoveExtensionIfUnsuitable(extensions.custom_border_color,
- VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME);
+ RemoveExtensionFeatureIfUnsuitable(extensions.custom_border_color, features.custom_border_color,
+ VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME);
// VK_EXT_depth_clip_control
extensions.depth_clip_control = features.depth_clip_control.depthClipControl;
- RemoveExtensionIfUnsuitable(extensions.depth_clip_control,
- VK_EXT_DEPTH_CLIP_CONTROL_EXTENSION_NAME);
+ RemoveExtensionFeatureIfUnsuitable(extensions.depth_clip_control, features.depth_clip_control,
+ VK_EXT_DEPTH_CLIP_CONTROL_EXTENSION_NAME);
// VK_EXT_extended_dynamic_state
extensions.extended_dynamic_state = features.extended_dynamic_state.extendedDynamicState;
- RemoveExtensionIfUnsuitable(extensions.extended_dynamic_state,
- VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME);
+ RemoveExtensionFeatureIfUnsuitable(extensions.extended_dynamic_state,
+ features.extended_dynamic_state,
+ VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME);
// VK_EXT_extended_dynamic_state2
extensions.extended_dynamic_state2 = features.extended_dynamic_state2.extendedDynamicState2;
- RemoveExtensionIfUnsuitable(extensions.extended_dynamic_state2,
- VK_EXT_EXTENDED_DYNAMIC_STATE_2_EXTENSION_NAME);
+ RemoveExtensionFeatureIfUnsuitable(extensions.extended_dynamic_state2,
+ features.extended_dynamic_state2,
+ VK_EXT_EXTENDED_DYNAMIC_STATE_2_EXTENSION_NAME);
// VK_EXT_extended_dynamic_state3
dynamic_state3_blending =
@@ -1048,35 +1072,38 @@ void Device::RemoveUnsuitableExtensions() {
extensions.extended_dynamic_state3 = dynamic_state3_blending || dynamic_state3_enables;
dynamic_state3_blending = dynamic_state3_blending && extensions.extended_dynamic_state3;
dynamic_state3_enables = dynamic_state3_enables && extensions.extended_dynamic_state3;
- RemoveExtensionIfUnsuitable(extensions.extended_dynamic_state3,
- VK_EXT_EXTENDED_DYNAMIC_STATE_3_EXTENSION_NAME);
+ RemoveExtensionFeatureIfUnsuitable(extensions.extended_dynamic_state3,
+ features.extended_dynamic_state3,
+ VK_EXT_EXTENDED_DYNAMIC_STATE_3_EXTENSION_NAME);
// VK_EXT_provoking_vertex
extensions.provoking_vertex =
features.provoking_vertex.provokingVertexLast &&
features.provoking_vertex.transformFeedbackPreservesProvokingVertex;
- RemoveExtensionIfUnsuitable(extensions.provoking_vertex,
- VK_EXT_PROVOKING_VERTEX_EXTENSION_NAME);
+ RemoveExtensionFeatureIfUnsuitable(extensions.provoking_vertex, features.provoking_vertex,
+ VK_EXT_PROVOKING_VERTEX_EXTENSION_NAME);
// VK_KHR_shader_atomic_int64
extensions.shader_atomic_int64 = features.shader_atomic_int64.shaderBufferInt64Atomics &&
features.shader_atomic_int64.shaderSharedInt64Atomics;
- RemoveExtensionIfUnsuitable(extensions.shader_atomic_int64,
- VK_KHR_SHADER_ATOMIC_INT64_EXTENSION_NAME);
+ RemoveExtensionFeatureIfUnsuitable(extensions.shader_atomic_int64, features.shader_atomic_int64,
+ VK_KHR_SHADER_ATOMIC_INT64_EXTENSION_NAME);
// VK_EXT_shader_demote_to_helper_invocation
extensions.shader_demote_to_helper_invocation =
features.shader_demote_to_helper_invocation.shaderDemoteToHelperInvocation;
- RemoveExtensionIfUnsuitable(extensions.shader_demote_to_helper_invocation,
- VK_EXT_SHADER_DEMOTE_TO_HELPER_INVOCATION_EXTENSION_NAME);
+ RemoveExtensionFeatureIfUnsuitable(extensions.shader_demote_to_helper_invocation,
+ features.shader_demote_to_helper_invocation,
+ VK_EXT_SHADER_DEMOTE_TO_HELPER_INVOCATION_EXTENSION_NAME);
// VK_EXT_subgroup_size_control
extensions.subgroup_size_control =
features.subgroup_size_control.subgroupSizeControl &&
properties.subgroup_size_control.minSubgroupSize <= GuestWarpSize &&
properties.subgroup_size_control.maxSubgroupSize >= GuestWarpSize;
- RemoveExtensionIfUnsuitable(extensions.subgroup_size_control,
- VK_EXT_SUBGROUP_SIZE_CONTROL_EXTENSION_NAME);
+ RemoveExtensionFeatureIfUnsuitable(extensions.subgroup_size_control,
+ features.subgroup_size_control,
+ VK_EXT_SUBGROUP_SIZE_CONTROL_EXTENSION_NAME);
// VK_EXT_transform_feedback
extensions.transform_feedback =
@@ -1086,24 +1113,27 @@ void Device::RemoveUnsuitableExtensions() {
properties.transform_feedback.maxTransformFeedbackBuffers > 0 &&
properties.transform_feedback.transformFeedbackQueries &&
properties.transform_feedback.transformFeedbackDraw;
- RemoveExtensionIfUnsuitable(extensions.transform_feedback,
- VK_EXT_TRANSFORM_FEEDBACK_EXTENSION_NAME);
+ RemoveExtensionFeatureIfUnsuitable(extensions.transform_feedback, features.transform_feedback,
+ VK_EXT_TRANSFORM_FEEDBACK_EXTENSION_NAME);
// VK_EXT_vertex_input_dynamic_state
extensions.vertex_input_dynamic_state =
features.vertex_input_dynamic_state.vertexInputDynamicState;
- RemoveExtensionIfUnsuitable(extensions.vertex_input_dynamic_state,
- VK_EXT_VERTEX_INPUT_DYNAMIC_STATE_EXTENSION_NAME);
+ RemoveExtensionFeatureIfUnsuitable(extensions.vertex_input_dynamic_state,
+ features.vertex_input_dynamic_state,
+ VK_EXT_VERTEX_INPUT_DYNAMIC_STATE_EXTENSION_NAME);
// VK_KHR_pipeline_executable_properties
if (Settings::values.renderer_shader_feedback.GetValue()) {
extensions.pipeline_executable_properties =
features.pipeline_executable_properties.pipelineExecutableInfo;
- RemoveExtensionIfUnsuitable(extensions.pipeline_executable_properties,
- VK_KHR_PIPELINE_EXECUTABLE_PROPERTIES_EXTENSION_NAME);
+ RemoveExtensionFeatureIfUnsuitable(extensions.pipeline_executable_properties,
+ features.pipeline_executable_properties,
+ VK_KHR_PIPELINE_EXECUTABLE_PROPERTIES_EXTENSION_NAME);
} else {
- extensions.pipeline_executable_properties = false;
- loaded_extensions.erase(VK_KHR_PIPELINE_EXECUTABLE_PROPERTIES_EXTENSION_NAME);
+ RemoveExtensionFeature(extensions.pipeline_executable_properties,
+ features.pipeline_executable_properties,
+ VK_KHR_PIPELINE_EXECUTABLE_PROPERTIES_EXTENSION_NAME);
}
// VK_KHR_workgroup_memory_explicit_layout
@@ -1113,8 +1143,9 @@ void Device::RemoveUnsuitableExtensions() {
features.workgroup_memory_explicit_layout.workgroupMemoryExplicitLayout8BitAccess &&
features.workgroup_memory_explicit_layout.workgroupMemoryExplicitLayout16BitAccess &&
features.workgroup_memory_explicit_layout.workgroupMemoryExplicitLayoutScalarBlockLayout;
- RemoveExtensionIfUnsuitable(extensions.workgroup_memory_explicit_layout,
- VK_KHR_WORKGROUP_MEMORY_EXPLICIT_LAYOUT_EXTENSION_NAME);
+ RemoveExtensionFeatureIfUnsuitable(extensions.workgroup_memory_explicit_layout,
+ features.workgroup_memory_explicit_layout,
+ VK_KHR_WORKGROUP_MEMORY_EXPLICIT_LAYOUT_EXTENSION_NAME);
}
void Device::SetupFamilies(VkSurfaceKHR surface) {
diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h
index be3ed45ff..488fdd313 100644
--- a/src/video_core/vulkan_common/vulkan_device.h
+++ b/src/video_core/vulkan_common/vulkan_device.h
@@ -20,7 +20,6 @@ VK_DEFINE_HANDLE(VmaAllocator)
// Vulkan version in the macro describes the minimum version required for feature availability.
// If the Vulkan version is lower than the required version, the named extension is required.
#define FOR_EACH_VK_FEATURE_1_1(FEATURE) \
- FEATURE(EXT, SubgroupSizeControl, SUBGROUP_SIZE_CONTROL, subgroup_size_control) \
FEATURE(KHR, 16BitStorage, 16BIT_STORAGE, bit16_storage) \
FEATURE(KHR, ShaderAtomicInt64, SHADER_ATOMIC_INT64, shader_atomic_int64) \
FEATURE(KHR, ShaderDrawParameters, SHADER_DRAW_PARAMETERS, shader_draw_parameters) \
@@ -36,7 +35,8 @@ VK_DEFINE_HANDLE(VmaAllocator)
#define FOR_EACH_VK_FEATURE_1_3(FEATURE) \
FEATURE(EXT, ShaderDemoteToHelperInvocation, SHADER_DEMOTE_TO_HELPER_INVOCATION, \
- shader_demote_to_helper_invocation)
+ shader_demote_to_helper_invocation) \
+ FEATURE(EXT, SubgroupSizeControl, SUBGROUP_SIZE_CONTROL, subgroup_size_control)
// Define all features which may be used by the implementation and require an extension here.
#define FOR_EACH_VK_FEATURE_EXT(FEATURE) \
@@ -639,8 +639,17 @@ private:
// Remove extensions which have incomplete feature support.
void RemoveUnsuitableExtensions();
+
+ void RemoveExtension(bool& extension, const std::string& extension_name);
void RemoveExtensionIfUnsuitable(bool is_suitable, const std::string& extension_name);
+ template <typename Feature>
+ void RemoveExtensionFeature(bool& extension, Feature& feature,
+ const std::string& extension_name);
+ template <typename Feature>
+ void RemoveExtensionFeatureIfUnsuitable(bool is_suitable, Feature& feature,
+ const std::string& extension_name);
+
/// Sets up queue families.
void SetupFamilies(VkSurfaceKHR surface);
generated by cgit v1.2.3 (git 2.25.1) at 2024-11-21 23:50:30 +0100