Merge pull request #394 from Devsh-Graphics-Programming/glsl_define_injection

GLSL define injection update for limits & features

Author: Erfan-Ahmadi

include/nbl/builtin/glsl/property_pool/copy.comp

@@ -1,5 +1,5 @@
 #version 440 core
-layout(local_size_x=NBL_LIMIT_MAX_OPTIMALLY_RESIDENT_WORKGROUP_INVOCATIONS) in;
+layout(local_size_x=NBL_GLSL_LIMIT_MAX_OPTIMALLY_RESIDENT_WORKGROUP_INVOCATIONS) in;
 
 #include "nbl/builtin/glsl/property_pool/transfer.glsl"
 
@@ -68,7 +68,7 @@ void main()
     const uint dstIndexOffset = transfer.dstIndexOffset-indicesToSkip;
     // set up loop
     const uint DWORDs = min(transfer.elementCount*propDWORDs,pc.endDWORD);
-    const uint dispatchSize = gl_NumWorkGroups[0]*NBL_LIMIT_MAX_OPTIMALLY_RESIDENT_WORKGROUP_INVOCATIONS;
+    const uint dispatchSize = gl_NumWorkGroups[0]*NBL_GLSL_LIMIT_MAX_OPTIMALLY_RESIDENT_WORKGROUP_INVOCATIONS;
     //
     const bool fill = bool(transfer.propertyDWORDsize_flags&(NBL_BUILTIN_PROPERTY_POOL_TRANSFER_EF_SRC_FILL<<flagsBitOffset));
     if (fill)

include/nbl/builtin/glsl/skinning/cache_update.comp

@@ -1,5 +1,5 @@
 #version 440 core
-layout(local_size_x=NBL_LIMIT_MAX_OPTIMALLY_RESIDENT_WORKGROUP_INVOCATIONS) in;
+layout(local_size_x=NBL_GLSL_LIMIT_MAX_OPTIMALLY_RESIDENT_WORKGROUP_INVOCATIONS) in;
 
 #include <nbl/builtin/glsl/skinning/cache_descriptor_set.glsl>
 #include <nbl/builtin/glsl/skinning/update_descriptor_set.glsl>
@@ -15,7 +15,7 @@ void main()
         return;
 
     const uint totalJointCount = jointCountInclPrefixSum[skinCount-1u];
-    const uint dispatchSize = NBL_LIMIT_MAX_OPTIMALLY_RESIDENT_WORKGROUP_INVOCATIONS*gl_NumWorkGroups[0];
+    const uint dispatchSize = NBL_GLSL_LIMIT_MAX_OPTIMALLY_RESIDENT_WORKGROUP_INVOCATIONS*gl_NumWorkGroups[0];
     for (uint jointIndex=gl_GlobalInvocationID.x; jointIndex<totalJointCount; jointIndex+=dispatchSize)
     {
         // TODO: implement via https://moderngpu.github.io/sortedsearch.html, find the upper and lower bounds of the workgroup, then go to town with sorted search

include/nbl/builtin/glsl/subgroup/basic_portability.glsl

@@ -17,9 +17,9 @@
 #define nbl_glsl_MinSubgroupSizeLog2 2
 #define nbl_glsl_MinSubgroupSize (0x1<<nbl_glsl_MinSubgroupSizeLog2)
 
-#ifdef NBL_IMPL_GL_NV_shader_thread_group
+#ifdef NBL_GLSL_IMPL_GL_NV_shader_thread_group
 	#define nbl_glsl_MaxSubgroupSizeLog2 5
-#elif defined(NBL_IMPL_GL_AMD_gcn_shader)||defined(NBL_IMPL_GL_ARB_shader_ballot)
+#elif defined(NBL_GLSL_IMPL_GL_AMD_gcn_shader)||defined(NBL_GLSL_IMPL_GL_ARB_shader_ballot)
 	#define nbl_glsl_MaxSubgroupSizeLog2 6
 #else
 	#define nbl_glsl_MaxSubgroupSizeLog2 7

include/nbl/builtin/glsl/transform_tree/global_transform_update_common.glsl

@@ -1,4 +1,4 @@
-layout(local_size_x=NBL_LIMIT_MAX_OPTIMALLY_RESIDENT_WORKGROUP_INVOCATIONS) in;
+layout(local_size_x=NBL_GLSL_LIMIT_MAX_OPTIMALLY_RESIDENT_WORKGROUP_INVOCATIONS) in;
 
 #define NBL_GLSL_TRANSFORM_TREE_POOL_NODE_RECOMPUTED_TIMESTAMP_DESCRIPTOR_QUALIFIERS coherent restrict
 #define NBL_GLSL_TRANSFORM_TREE_POOL_NODE_GLOBAL_TRANSFORM_DESCRIPTOR_QUALIFIERS coherent restrict
@@ -66,7 +66,7 @@ void main()
 #define NBL_GLSL_TRANSFORM_TREE_STACK_SIZE (NBL_GLSL_TRANSFORM_TREE_MAX_DEPTH-1)
     uint stack[NBL_GLSL_TRANSFORM_TREE_STACK_SIZE];
 
-    const uint dispatchSize = NBL_LIMIT_MAX_OPTIMALLY_RESIDENT_WORKGROUP_INVOCATIONS*gl_NumWorkGroups[0];
+    const uint dispatchSize = NBL_GLSL_LIMIT_MAX_OPTIMALLY_RESIDENT_WORKGROUP_INVOCATIONS*gl_NumWorkGroups[0];
     for (uint nodeID=gl_GlobalInvocationID.x; nodeID<nodesToUpdate.count; nodeID+=dispatchSize)
     {
         int stackPtr = 0;

include/nbl/builtin/glsl/transform_tree/relative_transform_update.comp

@@ -1,5 +1,5 @@
 #version 440 core
-layout(local_size_x=NBL_LIMIT_MAX_OPTIMALLY_RESIDENT_WORKGROUP_INVOCATIONS) in;
+layout(local_size_x=NBL_GLSL_LIMIT_MAX_OPTIMALLY_RESIDENT_WORKGROUP_INVOCATIONS) in;
 
 // disable descriptors we dont need
 #define NBL_GLSL_TRANSFORM_TREE_POOL_NODE_PARENT_DESCRIPTOR_DECLARED
@@ -12,7 +12,7 @@ layout(local_size_x=NBL_LIMIT_MAX_OPTIMALLY_RESIDENT_WORKGROUP_INVOCATIONS) in;
 
 void main()
 {
-    const uint dispatchSize = NBL_LIMIT_MAX_OPTIMALLY_RESIDENT_WORKGROUP_INVOCATIONS*gl_NumWorkGroups[0];
+    const uint dispatchSize = NBL_GLSL_LIMIT_MAX_OPTIMALLY_RESIDENT_WORKGROUP_INVOCATIONS*gl_NumWorkGroups[0];
     for (uint nodeID=gl_GlobalInvocationID.x; nodeID<relativeTransformModificationRequestRanges.rangeCount; nodeID+=dispatchSize)
     {
         const nbl_glsl_transform_tree_modification_request_range_t requestRange = relativeTransformModificationRequestRanges.data[nodeID];

include/nbl/builtin/glsl/utils/indirect_commands.glsl

@@ -33,7 +33,7 @@ struct nbl_glsl_DispatchIndirectCommand_t
 uint nbl_glsl_utils_computeOptimalPersistentWorkgroupDispatchSize(in uint elementCount, in uint workgroupSize, in uint workgroupSpinningProtection)
 {
     const uint infinitelyWideDeviceWGCount = (elementCount-1u)/(workgroupSize*workgroupSpinningProtection)+1u;
-    return min(infinitelyWideDeviceWGCount,NBL_LIMIT_MAX_RESIDENT_INVOCATIONS/NBL_LIMIT_MAX_OPTIMALLY_RESIDENT_WORKGROUP_INVOCATIONS);
+    return min(infinitelyWideDeviceWGCount,NBL_GLSL_LIMIT_MAX_RESIDENT_INVOCATIONS/NBL_GLSL_LIMIT_MAX_OPTIMALLY_RESIDENT_WORKGROUP_INVOCATIONS);
 }
 uint nbl_glsl_utils_computeOptimalPersistentWorkgroupDispatchSize(in uint elementCount, in uint workgroupSize)
 {

include/nbl/video/SPhysicalDeviceFeatures.h

@@ -12,6 +12,75 @@ enum E_SWAPCHAIN_MODE : uint32_t
     /* TODO: KHR_swapchain if SURFACE or DISPLAY flag present & KHR_display_swapchain if DISPLAY flag present */
 };
 
+//! [TODOS]:
+//! 
+//! ## LogicalDevice creation enabled features shouldn't necessarily equal the ones it reports as enabled (superset)
+//! 
+//! Basically what I'd imagine the usage of the API to be like.
+//! 
+//! **RARE: Creating a physical device with all advertised features/extensions:**
+//! ```cpp
+//! auto features = physicalDevice->getFeatures();
+//! 
+//! ILogicalDevice::SCreationParams params = {};
+//! params.queueParamsCount = ; // set queue stuff
+//! params.queueParams = ; // set queue stuff
+//! params.enabledFeatures = features;
+//! auto device = physicalDevice->createLogicalDevice(params);
+//! ```
+//! 
+//! **FREQUENT: Choosing a physical device with the features**
+//! ```cpp
+//! IPhysicalDevice::SRequiredProperties props = {}; // default initializes to apiVersion=1.1, deviceType = ET_UNKNOWN, pipelineCacheUUID = '\0', device UUID=`\0`, driverUUID=`\0`, deviceLUID=`\0`, deviceNodeMask= ~0u, driverID=UNKNOWN
+//! // example of particular config
+//! props.apiVersion = 1.2;
+//! props.deviceTypeMask = ~IPhysicalDevice::ET_CPU; // would be good to turn the enum into a mask
+//! props.driverIDMask = ~(EDI_AMD_PROPRIETARY|EDI_INTEL_PROPRIETARY_WINDOWS); // would be goot to turn the enum into a mask
+//! props.conformanceVersion = 1.2;
+//! 
+//! SDeviceFeatures requiredFeatures = {};
+//! requiredFeatures.rayQuery = true;
+//! 
+//! SDeviceLimits minimumLimits = {}; // would default initialize to worst possible values (small values for maximum sizes, large values for alignments, etc.)
+//! 
+//! // TODO: later add some stuff for requiring queue families, formats and minimum memory heap sizes
+//! 
+//! auto physicalDeviceCandidates = api->getCompatiblePhysicalDevices(props,requiredFeatures,minimumLimits,numSwapchains,supportedSwapchains,/*optional: would enforce tighter checks to actually accept compatibility, like formats, present modes and surface caps*/swapchainSupportDecider);
+//! if (physicalDeviceCandidates.empty())
+//! {
+//!     logError();
+//!     exit();
+//! }
+//! 
+//! // TODO: later iterate through candidate devices (fulfilling all the required criteria) to find the "best" one
+//! // std::sort(physicalDeviceCandidates.begin(),physicalDeviceCandidates.end(),SDefaultPhysicalDeviceOrder());
+//! auto physicalDevice = physicalDeviceCandidates.begin();
+//! assert(requiredFeatures < physicalDevice->getFeatures());
+//! assert(minimumLimits < physicalDevice->getLimits());
+//! 
+//! ILogicalDevice::SCreationParams params = {};
+//! params.queueParamsCount = ; // set queue stuff
+//! params.queueParams = ; // set queue stuff
+//! params.enabledFeatures = requiredFeatures;
+//! auto device = physicalDevice->createLogicalDevice(params);
+//! // this would be wrong, because during device creation we would enable additional features either due to:
+//! // - dependencies (like buffer address for raytracing)
+//! // - backend force-enabling them (like in OpenGL, where extensions are just enabled, you have no choice)
+//! // assert(requiredFeatures != device->getEnabledFeatures());
+//! assert(requiredFeatures < device->getEnabledFeatures());
+//! ```
+//! 
+//! ### `SDeviceFeatures` and `SDeviceLimits` should have a `operator<`
+//! 
+//! Basically to let us establish if features or limits are a superset of the requested.
+//! 
+//! If you need a `! = ` operator then define it as
+//! ```cpp
+//! inline bool operator!=(const& other) const
+//! {
+//!     return *this < other || other < *this;
+//! }
+//! ```
 struct SPhysicalDeviceFeatures
 {
     /* Vulkan 1.0 Core  */
@@ -212,9 +281,9 @@ struct SPhysicalDeviceFeatures
     /* Vulkan Extensions */
 
     /* RasterizationOrderAttachmentAccessFeaturesARM *//* VK_ARM_rasterization_order_attachment_access */
-    bool rasterizationOrderColorAttachmentAccess;
-    bool rasterizationOrderDepthAttachmentAccess;
-    bool rasterizationOrderStencilAttachmentAccess;
+    bool rasterizationOrderColorAttachmentAccess = false;
+    bool rasterizationOrderDepthAttachmentAccess = false;
+    bool rasterizationOrderStencilAttachmentAccess = false;
 
     // [DO NOT EXPOSE] Enables certain formats in Vulkan, we just enable them if available or else we need to make format support query functions in LogicalDevice as well
     /* 4444FormatsFeaturesEXT *//* VK_EXT_4444_formats */
@@ -421,7 +490,7 @@ struct SPhysicalDeviceFeatures
     bool rayTracingMotionBlurPipelineTraceRaysIndirect = false;
 
     /* CoverageReductionModeFeaturesNV *//* VK_NV_coverage_reduction_mode */
-    bool coverageReductionMode;
+    bool coverageReductionMode = false;
 
     /* DeviceGeneratedCommandsFeaturesNV *//* VK_NV_device_generated_commands */
     bool deviceGeneratedCommands = false;
@@ -643,7 +712,25 @@ struct SPhysicalDeviceFeatures
     // [TODO LATER] Won't expose for now, API changes necessary
     /* VK_AMD_texture_gather_bias_lod */
 
-    // [TODO LATER] when released in the SDK: https://github.com/Devsh-Graphics-Programming/Nabla/pull/357#discussion_r916899420
+    // [TODO LATER] when released in the SDK:
+    // -Support for `GLSL_EXT_ray_cull_mask`, lets call it `rayCullMask`
+    // - new pipeline stage and access masks but only in `KHR_synchronization2` which we don't use
+    // - two new acceleration structure query parameters
+    // - `rayTracingPipelineTraceRaysIndirect2` feature, same as `rayTracingPipelineTraceRaysIndirect` but with indirect SBTand dispatch dimensions
+    // 
+    // Lets have
+    // ```cpp
+    // bool accelerationStructureSizeAndBLASPointersQuery = false;
+    // 
+    // // Do not expose, we don't use KHR_synchronization2 yet
+    // //bool accelerationStructureCopyStageAndSBTAccessType;
+    // 
+    // bool rayCullMask = false;
+    // 
+    // bool rayTracingPipelineTraceRaysIndirectDimensionsAndSBT = false;
+    // ```
+    // 
+    // Lets enable `rayTracingMaintenance1`and `rayTracingPipelineTraceRaysIndirect2` whenever required by the above.
     /* VK_KHR_ray_tracing_maintenance1 *//* added in vk 1.3.213, the SDK isn't released yet at this moment :D */
 
     // [TODO LATER] requires extra API work to use
@@ -714,6 +801,9 @@ struct SPhysicalDeviceFeatures
     // [TODO] Always enable, expose as limit
     /* VK_KHR_spirv_1_4 */
 
+    // [TODO] handle with a single num
+    /* VK_KHR_display_swapchain */
+
     // [TODO LATER] (When it has documentation): Always enable, expose as limit
     /* VK_AMD_gpu_shader_half_float_fetch */
 
@@ -951,7 +1041,6 @@ struct SPhysicalDeviceFeatures
     // [TODO] Triage leftover extensions below    
 
     /* VK_NV_present_barrier */
-    /* VK_KHR_display_swapchain */
     /* VK_EXT_queue_family_foreign */
     /* VK_EXT_separate_stencil_usage */
     /* VK_KHR_create_renderpass2 */

include/nbl/video/SPhysicalDeviceLimits.h

@@ -182,7 +182,11 @@ struct SPhysicalDeviceLimits
     E_TRI_BOOLEAN shaderRoundingModeRTZFloat32 = ETB_DONT_KNOW;
     E_TRI_BOOLEAN shaderRoundingModeRTZFloat64 = ETB_DONT_KNOW;
 
-    //      or VK_EXT_descriptor_indexing:
+    // expose in 2 phases
+    // -Update After Bindand nonUniformEXT shader qualifier:
+    //      Descriptor Lifetime Tracking PR #345 will do this, cause I don't want to rewrite the tracking system again.
+    // -Actual Descriptor Indexing:
+    //      The whole 512k descriptor limits, runtime desc arrays, etc.will come later
     uint32_t maxUpdateAfterBindDescriptorsInAllPools = ~0u;
     bool shaderUniformBufferArrayNonUniformIndexingNative = false;
     bool shaderSampledImageArrayNonUniformIndexingNative = false;
@@ -211,9 +215,6 @@ struct SPhysicalDeviceLimits
     bool filterMinmaxSingleComponentFormats = false;
     bool filterMinmaxImageComponentMapping = false;
 
-    core::bitflag<asset::IImage::E_SAMPLE_COUNT_FLAGS> framebufferIntegerColorSampleCounts = asset::IImage::E_SAMPLE_COUNT_FLAGS(0u);
-
-
     /* Vulkan 1.3 Core  */
 
     //      or VK_EXT_subgroup_size_control:
@@ -380,26 +381,6 @@ struct SPhysicalDeviceLimits
     //uint32_t              maxDescriptorSetInlineUniformBlocks;
     //uint32_t              maxDescriptorSetUpdateAfterBindInlineUniformBlocks;
 
-    // [DO NOT EXPOSE] We will never expose this vendor specific meta-data (no new feature) to the user, but might use the extension to provide some cross platform meta-info in the Nabla section
-    /* ShaderCoreProperties2AMD *//* provided by VK_AMD_shader_core_properties2 */
-    //VkShaderCorePropertiesFlagsAMD    shaderCoreFeatures;
-    //uint32_t                          activeComputeUnitCount;
-    /* ShaderCorePropertiesAMD *//* provided by VK_AMD_shader_core_properties */
-    //uint32_t           shaderEngineCount;
-    //uint32_t           shaderArraysPerEngineCount;
-    //uint32_t           computeUnitsPerShaderArray;
-    //uint32_t           simdPerComputeUnit;
-    //uint32_t           wavefrontsPerSimd;
-    //uint32_t           wavefrontSize;
-    //uint32_t           sgprsPerSimd;
-    //uint32_t           minSgprAllocation;
-    //uint32_t           maxSgprAllocation;
-    //uint32_t           sgprAllocationGranularity;
-    //uint32_t           vgprsPerSimd;
-    //uint32_t           minVgprAllocation;
-    //uint32_t           maxVgprAllocation;
-    //uint32_t           vgprAllocationGranularity;
-
     // [DO NOT EXPOSE] right now, no idea if we'll ever expose and implement those but they'd all be false for OpenGL
     /* BlendOperationAdvancedPropertiesEXT *//* provided by VK_EXT_blend_operation_advanced */
     //uint32_t           advancedBlendMaxColorAttachments;
@@ -553,6 +534,14 @@ struct SPhysicalDeviceLimits
     //! uint32_t              maxVertexInputAttributeOffset;
     //! uint32_t              maxVertexInputBindingStride;
 
+    /*
+    - Spec states minimum supported value should be at least ESCF_1_BIT
+    - it might be different for each integer format, best way is to query your integer format from physical device using vkGetPhysicalDeviceImageFormatProperties and get the sampleCounts
+    https://www.khronos.org/registry/vulkan/specs/1.3-extensions/man/html/VkImageFormatProperties.html
+    */
+    // [DO NOT EXPOSE] because it might be different for every texture format and usage
+    // core::bitflag<asset::IImage::E_SAMPLE_COUNT_FLAGS> framebufferIntegerColorSampleCounts = asset::IImage::E_SAMPLE_COUNT_FLAGS(0u);
+
     /*  Always enabled, reported as limits */
     bool shaderOutputViewportIndex = false;     // ALIAS: VK_EXT_shader_viewport_index_layer
     bool shaderOutputLayer = false;             // ALIAS: VK_EXT_shader_viewport_index_layer
@@ -584,9 +573,9 @@ struct SPhysicalDeviceLimits
     bool postDepthCoverage = false; /* VK_EXT_post_depth_coverage */
     bool shaderStencilExport = false; /* VK_EXT_shader_stencil_export */
     bool decorateString = false; /* VK_GOOGLE_decorate_string */
-    bool externalFence = false; /* VK_KHR_external_fence_fd */ /* VK_KHR_external_fence_win32 */
-    bool externalMemory = false; /* VK_KHR_external_memory_fd */ /* VK_KHR_external_memory_win32 */
-    bool externalSemaphore = false; /* VK_KHR_external_semaphore_fd */ /* VK_KHR_external_semaphore_win32 */
+    bool externalFence = false; /* VK_KHR_external_fence_fd */ /* VK_KHR_external_fence_win32 */ // [TODO] requires instance extensions, add them
+    bool externalMemory = false; /* VK_KHR_external_memory_fd */ /* VK_KHR_external_memory_win32 */ // [TODO] requires instance extensions, add them
+    bool externalSemaphore = false; /* VK_KHR_external_semaphore_fd */ /* VK_KHR_external_semaphore_win32 */ // [TODO] requires instance extensions, add them
     bool shaderNonSemanticInfo = false; /* VK_KHR_shader_non_semantic_info */
     bool fragmentShaderBarycentric = false; /* VK_KHR_fragment_shader_barycentric */
     bool geometryShaderPassthrough = false; /* VK_NV_geometry_shader_passthrough */