v1.124

v1.124:

HIGHLIGHTS:

• dropped "mGPU" support (minor breaking change)
• improved VK performance

DETAILS:

• dropped "mGPU" support, which was incomplete in VK, was not implemented in D3D12
• added missing VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT flag
• VK: minor improvements
• resolved some TODOs
• refactoring

v1.123

v1.123:

HIGHLIGHTS:

• no interface changes!
• improvements and bug fixes

DETAILS:

• D3D11: fixed AMD AGS usage, which didn't match the latest version
• D3D11: setup UAV slot (63) for shader extensions (both NVAPI and AGS)
• D3D11: added an ability to create RS state via "CreateRasterizerState1" if "CreateRasterizerState2" is not available
• D3D11: fixed potential log spam in WaitForDrain if AGS is available
• D3D11: MultiDraw*Indirect calls don't use MultiDraw*Indirect extensions if drawNum = 1
• VK: added missing VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT flag if device address extension is supported
• VK: added support for handy VK_KHR_swapchain_mutable_format
• CMake: minor improvements

v1.122

v1.122:

HIGHLIGHTS:

• added missing indirect commands, clarified expected command signatures
• added minor missing functionality: better PSL, two sided stencil support
• blend constants moved to dynamic state for flexibility
• adding constants allowing to leave viewportNum and sampleMask zero-initialized
• reorganized GraphicsPipelineDesc to be more logical
• removed rasterizerDiscard which is not available in D3D
• explained LogicFunc, CompareFunc, StencilFunc, BlendFactor and BlendFunc, using useful bits of the spec

BREAKING CHANGES (last round of relatively big changes):

• multisample settings grouped into optional MultisampleDesc and moved out of RasterizationDesc
• InputAssemblyDesc split into InputAssemblyDesc and optional VertexInputDesc
• CmdDispatchMeshTasks renamed to CmdDrawMeshTasks because it's a draw command
• width, height, depth in DispatchRaysDesc renamed to x, y, z to be inline with other Dispatch commands
• depthBiasConstantFactor renamed to depthBias
• Draw* and Dispatch* parameters reorganized into structs to be inline with Indirect variants (same arguments, same order)

DETAILS:

• added CmdDrawMeshTasksIndirect and CmdDispatchRaysIndirect
• draw and dispatch parameters reorganized into structs: DrawDesc, DrawIndexedDesc, DispatchDesc, DrawMeshTasksDesc and DispatchRaysIndirectRays (the latter is a bit special)
• VK: added missing Programmable Sample Positions support
• VK: added missing line smoothing support
• added meaningful comments
• bug fixes & improvements
• updated .clang-format
• refactoring

v1.121

v1.121:

HIGHLIGHTS:

• NRI: improvements and bug fixes
• Helper: removed before state in (Texture/Buffer)UploadDesc to emphasize the initial raw state of a resource, we are uploading into
• D3D12: enhanced barriers: more parity with VK barriers
• WrapperVK: reworked DeviceCreationVKDesc and added enabledExtensions

MINOR (POTENTIALLY) BREAKING CHANGE:

For barriers containing Layout::PRESENT the best (only?) valid barriers are:
Rendering: before = {.access = NriAccessBits_UNKNOWN, .layout = NriLayout_UNKNOWN, .stages = NriStageBits_NONE}
Presenting: after = {.access = NriAccessBits_UNKNOWN, .layout = NriLayout_PRESENT, .stages = <needed stages>}

DETAILS:

• NRI: set of booleans converted to bit field in DeviceDesc
• NRI: exposed isMeshShaderPipelineStatsSupported in DeviceDesc
• NRI: clarified usage of PipelineStatisticsDesc
• NRI: removed DeviceDesc::isAPIValidationEnabled
• NRI: SPIRVBindingOffsets moved to DeviceCreation ext
• D3D12/VK: access bits of {UNKNOWN, PRESENT} barrier get silently patched to COMMON/MEMORY_READ for D3D12/VK
• D3D12: enhanced barriers: AccessBits::UNKNOWN = D3D12_BARRIER_ACCESS_NO_ACCESS, Layout::UNKNOWN = D3D12_BARRIER_LAYOUT_UNDEFINED
• D3D12: D3D12_QUERY_DATA_PIPELINE_STATISTICS1 usage moved under NRI_USE_AGILITY_SDK macro
• VK: rewritten initialization (with bug fixes)
• Helper: removed "before" state in (Texture/Buffer)UploadDesc
• WrapperVK: reworked DeviceCreationVKDesc and added enabledExtensions
• deleted outdated programming guide
• updated NVAPI
• updated README
• polishing

v1.120

v1.120:

HIGHLIGHTS:

• NRI: bug fixes and improvements for Queries
• NRI: explicitly exposed isRaytracingSupported and isMeshShaderSupported in DeviceDesc

DETAILS:

• NRI: clarified query usage
• NRI: explicitly exposed isRaytracingSupported and isMeshShaderSupported in DeviceDesc
• NRI: fixed "unreferenced variable" warnings
• VK: fixed not working queries
• D3D11/D3D12/VK: reduced code entropy for queries

v1.119

v1.119:

HIGHLIGHTS:

• NRI: clarified usage in the main header
• NRI: explicitly exposed DepthAttachmentDesc::boundsTest enablement (expects CmdSetDepthBounds)
• NRI: removed StencilAttachmentDesc::reference (use CmdSetStencilReference instead)
• NRI: bug fixes and improvements
• Helper: UploadData improvements

DETAILS:

• NRI: added comments in NRI.h clarifying expected behavior of the dynamic state
• NRI: explicitly exposed depth bound test enablement
• NRI: removed StencilAttachmentDesc::reference (used only by VK, was a leftover)
• NRI: CHECK is now a wrapper around assert
• D3D11/D3D12: resolved issues around "depth bounds test"
• D3D12/VK: fixed mesh shader props in device desc
• D3D12: fixed referencing of temp (deallocated) memory on the stack during graphics pipeline creation
• VK: added printing of supported extensions (as INFO)
• Helper: UploadData improvements: mipNum and arraySize can be implicitly taken from the texture
• Helper: UploadData improvements: added missed optional before state (also added StageBits)

v1.118

v1.118:

BREAKING CHANGES!

HIGHLIGHTS:

• NRI: generalized barriers (D3D12 "enhanced barriers" and VK_KHR_synchronization2)
• NRI: refactoring to preserve interface health and agility
• NRI: ShaderStage, PipelineLayoutShaderStageBits and Dependency generalized to StageBits
• NRI: improved NRIDescs.h layout, added "pragma regions"
• D3D11/D3D12/VK: improvements and bug fixes

DETAILS:

• NRI: refactoring (see Crib Sheet)
• VK: barriers use VK_KHR_synchronization2
• VK: bug fixes reported by updated validation
• VK: improved extensions handling
• D3D11: exposed isNVAPILoaded in DeviceCreationD3D11Desc
• D3D11: fixed memory leak caused by wrapped device
• D3D11: fixed potentially wrong usage of "UAV overlap" extension
• D3D12: enhanced barriers support with soft fallback to legacy barriers
• D3D12: utilized "latest" interfaces where possible
• D3D11/D3D12: more flexible interface querying
• D3D11/D3D12: fixed potentially wrong swap chain interface querying
• D3D11/D3D12/VK: depth bounds get reset in SetPipeline per VK spec
• VALIDATION: improvements to reflect the changes
• Helper: removed ChangeResourceStates
• Helper: optimized barriers in UploadData

CRIB SHEET:

• CmdPipelineBarrier => CmdBarrier
• Dependency => StageBits
• PipelineLayoutShaderStageBits => StageBits
• ShaderStage => StageBits
• TextureTransitionBarrierDesc => TextureBararierDesc
• BufferTransitionBarrierDesc => BufferBararierDesc
• TransitionBarrierDesc => BarrierGroupDesc
• TextureLayout => Layout
• PipelineLayoutDesc::stageMask => shaderStages
• DescriptorRangeDesc::visibility => shaderStages
• ComputePipelineDesc::computeShader => shader
• GraphicsPipelineDesc::shaderStages => shaders
• GraphicsPipelineDesc::shaderStageNum => shaderNum
• ShaderLibrary::shaderDescs => shaders
• TextureTransitionBarrierDesc::nextState => after
• BufferTransitionBarrierDesc::nextState => after
• BufferUploadDesc::nextAccess => after
• TextureUploadDesc::nextAccess => after
• TextureTransitionBarrierDesc::prevState => before
• BufferTransitionBarrierDesc::prevState => before
• BufferUploadDesc::prevAccess => before
• ShaderStage::VERTEX => VERTEX_SHADER
• ShaderStage::TESS_CONTROL => TESS_CONTROL_SHADER
• ShaderStage::TESS_EVALUATION => TESS_EVALUATION_SHADER
• ShaderStage::GEOMETRY => GEOMETRY_SHADER
• ShaderStage::MESH_CONTROL => MESH_CONTROL_SHADER
• ShaderStage::MESH_EVALUATION => MESH_EVALUATION_SHADER
• ShaderStage::FRAGMENT => FRAGMENT_SHADER
• ShaderStage::COMPUTE => COMPUTE_SHADER
• ShaderStage::RAYGEN => RAYGEN_SHADER
• ShaderStage::MISS => MISS_SHADER
• ShaderStage::INTERSECTION => INTERSECTION_SHADER
• ShaderStage::CLOSEST_HIT => CLOSEST_HIT_SHADER
• ShaderStage::ANY_HIT => ANY_HIT_SHADER
• ShaderStage::CALLABLE => CALLABLE_SHADER
• TextureLayout::GENERAL => SHADER_RESOURCE_STORAGE

v1.117

v1.117:

• VK: added ability to skip VK ray tracing initialization during device creation to save memory
• NRI: removed experimental StreamLine integration
• D3D11: minor renaming in DeviceCreationDesc
• D3D12: minor code cleanup
• updated docs and CMake

v1.116

v1.116:

HIGHLIGHTS:

• fixes and improvements
• merged PRs from users

DETAILS:

• NRI: retrieve adapter desc as early as possible to improve message reporting
• NRI: fixed GetDisplayDesc behavior if a window is outside of all displays
• NRI: improved error reporting
• VK: added missing forward declaration in wrapper ext
• VK: improved memory type selection
• VK: fixed wrapper compatibility with C interface

v1.115

v1.115:

HIGHLIGHTS:

• NRI: added MemoryLocation::DEVICE_UPLOAD to get access to GPU_UPLOAD or (Resizable) BAR heap with silent fallback to HOST_UPLOAD
• Cmake: explicit specification of required backends
• Cmake: optional Agility SDK support (needed for D3D12 to get access to recent features, like ReBAR)
• fixes and improvements

DETAILS:

• NRI: added MemoryLocation::DEVICE_UPLOAD (ReBAR)
• NRI: fixed compilation on Android due to missing _mm_pause
• VK: improved "memoryType" selection
• VK: improved memory heaps info reporting
• Cmake improvements
• updated README