diff options
Diffstat (limited to 'src/video_core/vulkan_common')
| -rw-r--r-- | src/video_core/vulkan_common/vulkan_device.cpp | 23 | ||||
| -rw-r--r-- | src/video_core/vulkan_common/vulkan_device.h | 85 | ||||
| -rw-r--r-- | src/video_core/vulkan_common/vulkan_wrapper.cpp | 5 | ||||
| -rw-r--r-- | src/video_core/vulkan_common/vulkan_wrapper.h | 40 |
4 files changed, 126 insertions, 27 deletions
diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp index 617417040..3960b135a 100644 --- a/src/video_core/vulkan_common/vulkan_device.cpp +++ b/src/video_core/vulkan_common/vulkan_device.cpp @@ -76,6 +76,11 @@ constexpr std::array VK_FORMAT_R32G32B32_SFLOAT{ VK_FORMAT_UNDEFINED, }; +constexpr std::array VK_FORMAT_A4B4G4R4_UNORM_PACK16{ + VK_FORMAT_R4G4B4A4_UNORM_PACK16, + VK_FORMAT_UNDEFINED, +}; + } // namespace Alternatives enum class NvidiaArchitecture { @@ -110,6 +115,8 @@ constexpr const VkFormat* GetFormatAlternatives(VkFormat format) { return Alternatives::R8G8B8_SSCALED.data(); case VK_FORMAT_R32G32B32_SFLOAT: return Alternatives::VK_FORMAT_R32G32B32_SFLOAT.data(); + case VK_FORMAT_A4B4G4R4_UNORM_PACK16_EXT: + return Alternatives::VK_FORMAT_A4B4G4R4_UNORM_PACK16.data(); default: return nullptr; } @@ -238,6 +245,7 @@ std::unordered_map<VkFormat, VkFormatProperties> GetFormatProperties(vk::Physica VK_FORMAT_R32_SINT, VK_FORMAT_R32_UINT, VK_FORMAT_R4G4B4A4_UNORM_PACK16, + VK_FORMAT_A4B4G4R4_UNORM_PACK16_EXT, VK_FORMAT_R4G4_UNORM_PACK8, VK_FORMAT_R5G5B5A1_UNORM_PACK16, VK_FORMAT_R5G6B5_UNORM_PACK16, @@ -420,7 +428,8 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR first_next = &diagnostics_nv; } - is_blit_depth_stencil_supported = TestDepthStencilBlits(); + is_blit_depth24_stencil8_supported = TestDepthStencilBlits(VK_FORMAT_D24_UNORM_S8_UINT); + is_blit_depth32_stencil8_supported = TestDepthStencilBlits(VK_FORMAT_D32_SFLOAT_S8_UINT); is_optimal_astc_supported = ComputeIsOptimalAstcSupported(); is_warp_potentially_bigger = !extensions.subgroup_size_control || properties.subgroup_size_control.maxSubgroupSize > GuestWarpSize; @@ -774,14 +783,13 @@ bool Device::ComputeIsOptimalAstcSupported() const { return true; } -bool Device::TestDepthStencilBlits() const { +bool Device::TestDepthStencilBlits(VkFormat format) const { static constexpr VkFormatFeatureFlags required_features = VK_FORMAT_FEATURE_BLIT_SRC_BIT | VK_FORMAT_FEATURE_BLIT_DST_BIT; const auto test_features = [](VkFormatProperties props) { return (props.optimalTilingFeatures & required_features) == required_features; }; - return test_features(format_properties.at(VK_FORMAT_D32_SFLOAT_S8_UINT)) && - test_features(format_properties.at(VK_FORMAT_D24_UNORM_S8_UINT)); + return test_features(format_properties.at(format)); } bool Device::IsFormatSupported(VkFormat wanted_format, VkFormatFeatureFlags wanted_usage, @@ -1051,6 +1059,13 @@ void Device::RemoveUnsuitableExtensions() { RemoveExtensionFeatureIfUnsuitable(extensions.custom_border_color, features.custom_border_color, VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME); + // VK_EXT_depth_bias_control + extensions.depth_bias_control = + features.depth_bias_control.depthBiasControl && + features.depth_bias_control.leastRepresentableValueForceUnormRepresentation; + RemoveExtensionFeatureIfUnsuitable(extensions.depth_bias_control, features.depth_bias_control, + VK_EXT_DEPTH_BIAS_CONTROL_EXTENSION_NAME); + // VK_EXT_depth_clip_control extensions.depth_clip_control = features.depth_clip_control.depthClipControl; RemoveExtensionFeatureIfUnsuitable(extensions.depth_clip_control, features.depth_clip_control, diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h index 488fdd313..9be612392 100644 --- a/src/video_core/vulkan_common/vulkan_device.h +++ b/src/video_core/vulkan_common/vulkan_device.h @@ -41,10 +41,12 @@ VK_DEFINE_HANDLE(VmaAllocator) // Define all features which may be used by the implementation and require an extension here. #define FOR_EACH_VK_FEATURE_EXT(FEATURE) \ FEATURE(EXT, CustomBorderColor, CUSTOM_BORDER_COLOR, custom_border_color) \ + FEATURE(EXT, DepthBiasControl, DEPTH_BIAS_CONTROL, depth_bias_control) \ FEATURE(EXT, DepthClipControl, DEPTH_CLIP_CONTROL, depth_clip_control) \ FEATURE(EXT, ExtendedDynamicState, EXTENDED_DYNAMIC_STATE, extended_dynamic_state) \ FEATURE(EXT, ExtendedDynamicState2, EXTENDED_DYNAMIC_STATE_2, extended_dynamic_state2) \ FEATURE(EXT, ExtendedDynamicState3, EXTENDED_DYNAMIC_STATE_3, extended_dynamic_state3) \ + FEATURE(EXT, 4444Formats, 4444_FORMATS, format_a4b4g4r4) \ FEATURE(EXT, IndexTypeUint8, INDEX_TYPE_UINT8, index_type_uint8) \ FEATURE(EXT, LineRasterization, LINE_RASTERIZATION, line_rasterization) \ FEATURE(EXT, PrimitiveTopologyListRestart, PRIMITIVE_TOPOLOGY_LIST_RESTART, \ @@ -60,6 +62,7 @@ VK_DEFINE_HANDLE(VmaAllocator) // Define miscellaneous extensions which may be used by the implementation here. #define FOR_EACH_VK_EXTENSION(EXTENSION) \ + EXTENSION(EXT, CONDITIONAL_RENDERING, conditional_rendering) \ EXTENSION(EXT, CONSERVATIVE_RASTERIZATION, conservative_rasterization) \ EXTENSION(EXT, DEPTH_RANGE_UNRESTRICTED, depth_range_unrestricted) \ EXTENSION(EXT, MEMORY_BUDGET, memory_budget) \ @@ -92,11 +95,14 @@ VK_DEFINE_HANDLE(VmaAllocator) // Define extensions where the absence of the extension may result in a degraded experience. #define FOR_EACH_VK_RECOMMENDED_EXTENSION(EXTENSION_NAME) \ + EXTENSION_NAME(VK_EXT_CONDITIONAL_RENDERING_EXTENSION_NAME) \ EXTENSION_NAME(VK_EXT_CONSERVATIVE_RASTERIZATION_EXTENSION_NAME) \ + EXTENSION_NAME(VK_EXT_DEPTH_BIAS_CONTROL_EXTENSION_NAME) \ EXTENSION_NAME(VK_EXT_DEPTH_RANGE_UNRESTRICTED_EXTENSION_NAME) \ EXTENSION_NAME(VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME) \ EXTENSION_NAME(VK_EXT_EXTENDED_DYNAMIC_STATE_2_EXTENSION_NAME) \ EXTENSION_NAME(VK_EXT_EXTENDED_DYNAMIC_STATE_3_EXTENSION_NAME) \ + EXTENSION_NAME(VK_EXT_4444_FORMATS_EXTENSION_NAME) \ EXTENSION_NAME(VK_EXT_LINE_RASTERIZATION_EXTENSION_NAME) \ EXTENSION_NAME(VK_EXT_ROBUSTNESS_2_EXTENSION_NAME) \ EXTENSION_NAME(VK_EXT_VERTEX_INPUT_DYNAMIC_STATE_EXTENSION_NAME) \ @@ -143,7 +149,11 @@ VK_DEFINE_HANDLE(VmaAllocator) // Define features where the absence of the feature may result in a degraded experience. #define FOR_EACH_VK_RECOMMENDED_FEATURE(FEATURE_NAME) \ FEATURE_NAME(custom_border_color, customBorderColors) \ + FEATURE_NAME(depth_bias_control, depthBiasControl) \ + FEATURE_NAME(depth_bias_control, leastRepresentableValueForceUnormRepresentation) \ + FEATURE_NAME(depth_bias_control, depthBiasExact) \ FEATURE_NAME(extended_dynamic_state, extendedDynamicState) \ + FEATURE_NAME(format_a4b4g4r4, formatA4B4G4R4) \ FEATURE_NAME(index_type_uint8, indexTypeUint8) \ FEATURE_NAME(primitive_topology_list_restart, primitiveTopologyListRestart) \ FEATURE_NAME(provoking_vertex, provokingVertexLast) \ @@ -319,6 +329,11 @@ public: return features.shader_float16_int8.shaderInt8; } + /// Returns true if the device supports binding multisample images as storage images. + bool IsStorageImageMultisampleSupported() const { + return features.features.shaderStorageImageMultisample; + } + /// Returns true if the device warp size can potentially be bigger than guest's warp size. bool IsWarpSizePotentiallyBiggerThanGuest() const { return is_warp_potentially_bigger; @@ -359,9 +374,14 @@ public: return features.features.depthBounds; } - /// Returns true when blitting from and to depth stencil images is supported. - bool IsBlitDepthStencilSupported() const { - return is_blit_depth_stencil_supported; + /// Returns true when blitting from and to D24S8 images is supported. + bool IsBlitDepth24Stencil8Supported() const { + return is_blit_depth24_stencil8_supported; + } + + /// Returns true when blitting from and to D32S8 images is supported. + bool IsBlitDepth32Stencil8Supported() const { + return is_blit_depth32_stencil8_supported; } /// Returns true if the device supports VK_NV_viewport_swizzle. @@ -449,6 +469,11 @@ public: return extensions.depth_clip_control; } + /// Returns true if the device supports VK_EXT_depth_bias_control. + bool IsExtDepthBiasControlSupported() const { + return extensions.depth_bias_control; + } + /// Returns true if the device supports VK_EXT_shader_viewport_index_layer. bool IsExtShaderViewportIndexLayerSupported() const { return extensions.shader_viewport_index_layer; @@ -488,6 +513,11 @@ public: return extensions.extended_dynamic_state3; } + /// Returns true if the device supports VK_EXT_4444_formats. + bool IsExt4444FormatsSupported() const { + return features.format_a4b4g4r4.formatA4B4G4R4; + } + /// Returns true if the device supports VK_EXT_extended_dynamic_state3. bool IsExtExtendedDynamicState3BlendingSupported() const { return dynamic_state3_blending; @@ -528,6 +558,10 @@ public: return extensions.shader_atomic_int64; } + bool IsExtConditionalRendering() const { + return extensions.conditional_rendering; + } + bool HasTimelineSemaphore() const; /// Returns the minimum supported version of SPIR-V. @@ -600,6 +634,10 @@ public: return features.robustness2.nullDescriptor; } + bool HasExactDepthBiasControl() const { + return features.depth_bias_control.depthBiasExact; + } + u32 GetMaxVertexInputAttributes() const { return properties.properties.limits.maxVertexInputAttributes; } @@ -666,7 +704,7 @@ private: bool ComputeIsOptimalAstcSupported() const; /// Returns true if the device natively supports blitting depth stencil images. - bool TestDepthStencilBlits() const; + bool TestDepthStencilBlits(VkFormat format) const; private: VkInstance instance; ///< Vulkan instance. @@ -730,25 +768,26 @@ private: VkPhysicalDeviceProperties2 properties2{}; // Misc features - bool is_optimal_astc_supported{}; ///< Support for all guest ASTC formats. - bool is_blit_depth_stencil_supported{}; ///< Support for blitting from and to depth stencil. - bool is_warp_potentially_bigger{}; ///< Host warp size can be bigger than guest. - bool is_integrated{}; ///< Is GPU an iGPU. - bool is_virtual{}; ///< Is GPU a virtual GPU. - bool is_non_gpu{}; ///< Is SoftwareRasterizer, FPGA, non-GPU device. - bool has_broken_compute{}; ///< Compute shaders can cause crashes - bool has_broken_cube_compatibility{}; ///< Has broken cube compatibility bit - bool has_renderdoc{}; ///< Has RenderDoc attached - bool has_nsight_graphics{}; ///< Has Nsight Graphics attached - bool supports_d24_depth{}; ///< Supports D24 depth buffers. - bool cant_blit_msaa{}; ///< Does not support MSAA<->MSAA blitting. - bool must_emulate_scaled_formats{}; ///< Requires scaled vertex format emulation - bool must_emulate_bgr565{}; ///< Emulates BGR565 by swizzling RGB565 format. - bool dynamic_state3_blending{}; ///< Has all blending features of dynamic_state3. - bool dynamic_state3_enables{}; ///< Has all enables features of dynamic_state3. - bool supports_conditional_barriers{}; ///< Allows barriers in conditional control flow. - u64 device_access_memory{}; ///< Total size of device local memory in bytes. - u32 sets_per_pool{}; ///< Sets per Description Pool + bool is_optimal_astc_supported{}; ///< Support for all guest ASTC formats. + bool is_blit_depth24_stencil8_supported{}; ///< Support for blitting from and to D24S8. + bool is_blit_depth32_stencil8_supported{}; ///< Support for blitting from and to D32S8. + bool is_warp_potentially_bigger{}; ///< Host warp size can be bigger than guest. + bool is_integrated{}; ///< Is GPU an iGPU. + bool is_virtual{}; ///< Is GPU a virtual GPU. + bool is_non_gpu{}; ///< Is SoftwareRasterizer, FPGA, non-GPU device. + bool has_broken_compute{}; ///< Compute shaders can cause crashes + bool has_broken_cube_compatibility{}; ///< Has broken cube compatibility bit + bool has_renderdoc{}; ///< Has RenderDoc attached + bool has_nsight_graphics{}; ///< Has Nsight Graphics attached + bool supports_d24_depth{}; ///< Supports D24 depth buffers. + bool cant_blit_msaa{}; ///< Does not support MSAA<->MSAA blitting. + bool must_emulate_scaled_formats{}; ///< Requires scaled vertex format emulation + bool must_emulate_bgr565{}; ///< Emulates BGR565 by swizzling RGB565 format. + bool dynamic_state3_blending{}; ///< Has all blending features of dynamic_state3. + bool dynamic_state3_enables{}; ///< Has all enables features of dynamic_state3. + bool supports_conditional_barriers{}; ///< Allows barriers in conditional control flow. + u64 device_access_memory{}; ///< Total size of device local memory in bytes. + u32 sets_per_pool{}; ///< Sets per Description Pool // Telemetry parameters std::set<std::string, std::less<>> supported_extensions; ///< Reported Vulkan extensions. diff --git a/src/video_core/vulkan_common/vulkan_wrapper.cpp b/src/video_core/vulkan_common/vulkan_wrapper.cpp index c3f388d89..2f3254a97 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.cpp +++ b/src/video_core/vulkan_common/vulkan_wrapper.cpp @@ -75,6 +75,7 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { X(vkBeginCommandBuffer); X(vkBindBufferMemory); X(vkBindImageMemory); + X(vkCmdBeginConditionalRenderingEXT); X(vkCmdBeginQuery); X(vkCmdBeginRenderPass); X(vkCmdBeginTransformFeedbackEXT); @@ -91,6 +92,7 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { X(vkCmdCopyBufferToImage); X(vkCmdCopyImage); X(vkCmdCopyImageToBuffer); + X(vkCmdCopyQueryPoolResults); X(vkCmdDispatch); X(vkCmdDispatchIndirect); X(vkCmdDraw); @@ -99,6 +101,8 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { X(vkCmdDrawIndexedIndirect); X(vkCmdDrawIndirectCount); X(vkCmdDrawIndexedIndirectCount); + X(vkCmdDrawIndirectByteCountEXT); + X(vkCmdEndConditionalRenderingEXT); X(vkCmdEndQuery); X(vkCmdEndRenderPass); X(vkCmdEndTransformFeedbackEXT); @@ -109,6 +113,7 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept { X(vkCmdPushDescriptorSetWithTemplateKHR); X(vkCmdSetBlendConstants); X(vkCmdSetDepthBias); + X(vkCmdSetDepthBias2EXT); X(vkCmdSetDepthBounds); X(vkCmdSetEvent); X(vkCmdSetScissor); diff --git a/src/video_core/vulkan_common/vulkan_wrapper.h b/src/video_core/vulkan_common/vulkan_wrapper.h index 049fa8038..1e3c0fa64 100644 --- a/src/video_core/vulkan_common/vulkan_wrapper.h +++ b/src/video_core/vulkan_common/vulkan_wrapper.h @@ -185,6 +185,7 @@ struct DeviceDispatch : InstanceDispatch { PFN_vkBeginCommandBuffer vkBeginCommandBuffer{}; PFN_vkBindBufferMemory vkBindBufferMemory{}; PFN_vkBindImageMemory vkBindImageMemory{}; + PFN_vkCmdBeginConditionalRenderingEXT vkCmdBeginConditionalRenderingEXT{}; PFN_vkCmdBeginDebugUtilsLabelEXT vkCmdBeginDebugUtilsLabelEXT{}; PFN_vkCmdBeginQuery vkCmdBeginQuery{}; PFN_vkCmdBeginRenderPass vkCmdBeginRenderPass{}; @@ -202,6 +203,7 @@ struct DeviceDispatch : InstanceDispatch { PFN_vkCmdCopyBufferToImage vkCmdCopyBufferToImage{}; PFN_vkCmdCopyImage vkCmdCopyImage{}; PFN_vkCmdCopyImageToBuffer vkCmdCopyImageToBuffer{}; + PFN_vkCmdCopyQueryPoolResults vkCmdCopyQueryPoolResults{}; PFN_vkCmdDispatch vkCmdDispatch{}; PFN_vkCmdDispatchIndirect vkCmdDispatchIndirect{}; PFN_vkCmdDraw vkCmdDraw{}; @@ -210,6 +212,8 @@ struct DeviceDispatch : InstanceDispatch { PFN_vkCmdDrawIndexedIndirect vkCmdDrawIndexedIndirect{}; PFN_vkCmdDrawIndirectCount vkCmdDrawIndirectCount{}; PFN_vkCmdDrawIndexedIndirectCount vkCmdDrawIndexedIndirectCount{}; + PFN_vkCmdDrawIndirectByteCountEXT vkCmdDrawIndirectByteCountEXT{}; + PFN_vkCmdEndConditionalRenderingEXT vkCmdEndConditionalRenderingEXT{}; PFN_vkCmdEndDebugUtilsLabelEXT vkCmdEndDebugUtilsLabelEXT{}; PFN_vkCmdEndQuery vkCmdEndQuery{}; PFN_vkCmdEndRenderPass vkCmdEndRenderPass{}; @@ -222,6 +226,7 @@ struct DeviceDispatch : InstanceDispatch { PFN_vkCmdSetBlendConstants vkCmdSetBlendConstants{}; PFN_vkCmdSetCullModeEXT vkCmdSetCullModeEXT{}; PFN_vkCmdSetDepthBias vkCmdSetDepthBias{}; + PFN_vkCmdSetDepthBias2EXT vkCmdSetDepthBias2EXT{}; PFN_vkCmdSetDepthBounds vkCmdSetDepthBounds{}; PFN_vkCmdSetDepthBoundsTestEnableEXT vkCmdSetDepthBoundsTestEnableEXT{}; PFN_vkCmdSetDepthCompareOpEXT vkCmdSetDepthCompareOpEXT{}; @@ -1182,6 +1187,13 @@ public: count_offset, draw_count, stride); } + void DrawIndirectByteCountEXT(u32 instance_count, u32 first_instance, VkBuffer counter_buffer, + VkDeviceSize counter_buffer_offset, u32 counter_offset, + u32 stride) { + dld->vkCmdDrawIndirectByteCountEXT(handle, instance_count, first_instance, counter_buffer, + counter_buffer_offset, counter_offset, stride); + } + void ClearAttachments(Span<VkClearAttachment> attachments, Span<VkClearRect> rects) const noexcept { dld->vkCmdClearAttachments(handle, attachments.size(), attachments.data(), rects.size(), @@ -1270,6 +1282,13 @@ public: regions.data()); } + void CopyQueryPoolResults(VkQueryPool query_pool, u32 first_query, u32 query_count, + VkBuffer dst_buffer, VkDeviceSize dst_offset, VkDeviceSize stride, + VkQueryResultFlags flags) const noexcept { + dld->vkCmdCopyQueryPoolResults(handle, query_pool, first_query, query_count, dst_buffer, + dst_offset, stride, flags); + } + void FillBuffer(VkBuffer dst_buffer, VkDeviceSize dst_offset, VkDeviceSize size, u32 data) const noexcept { dld->vkCmdFillBuffer(handle, dst_buffer, dst_offset, size, data); @@ -1315,6 +1334,18 @@ public: dld->vkCmdSetDepthBias(handle, constant_factor, clamp, slope_factor); } + void SetDepthBias(float constant_factor, float clamp, float slope_factor, + VkDepthBiasRepresentationInfoEXT* extra) const noexcept { + VkDepthBiasInfoEXT info{ + .sType = VK_STRUCTURE_TYPE_DEPTH_BIAS_INFO_EXT, + .pNext = extra, + .depthBiasConstantFactor = constant_factor, + .depthBiasClamp = clamp, + .depthBiasSlopeFactor = slope_factor, + }; + dld->vkCmdSetDepthBias2EXT(handle, &info); + } + void SetDepthBounds(float min_depth_bounds, float max_depth_bounds) const noexcept { dld->vkCmdSetDepthBounds(handle, min_depth_bounds, max_depth_bounds); } @@ -1448,6 +1479,15 @@ public: counter_buffers, counter_buffer_offsets); } + void BeginConditionalRenderingEXT( + const VkConditionalRenderingBeginInfoEXT& info) const noexcept { + dld->vkCmdBeginConditionalRenderingEXT(handle, &info); + } + + void EndConditionalRenderingEXT() const noexcept { + dld->vkCmdEndConditionalRenderingEXT(handle); + } + void BeginDebugUtilsLabelEXT(const char* label, std::span<float, 4> color) const noexcept { const VkDebugUtilsLabelEXT label_info{ .sType = VK_STRUCTURE_TYPE_DEBUG_UTILS_LABEL_EXT, |