blob: 809d94cc3c8fbaf779248b3dff7c3bb6fdc1ddfd [file] [log] [blame]
/* Copyright (c) 2015-2019 The Khronos Group Inc.
* Copyright (c) 2015-2019 Valve Corporation
* Copyright (c) 2015-2019 LunarG, Inc.
* Copyright (C) 2015-2019 Google Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
* Author: Courtney Goeltzenleuchter <courtneygo@google.com>
* Author: Tobin Ehlis <tobine@google.com>
* Author: Chris Forbes <chrisf@ijw.co.nz>
* Author: Mark Lobodzinski <mark@lunarg.com>
* Author: Dave Houlton <daveh@lunarg.com>
* Author: John Zulauf <jzulauf@lunarg.com>
*/
#ifndef CORE_VALIDATION_TYPES_H_
#define CORE_VALIDATION_TYPES_H_
#include "cast_utils.h"
#include "hash_vk_types.h"
#include "sparse_containers.h"
#include "vk_safe_struct.h"
#include "vulkan/vulkan.h"
#include "vk_layer_logging.h"
#include "vk_object_types.h"
#include "vk_extension_helper.h"
#include "vk_typemap_helper.h"
#include "convert_to_renderpass2.h"
#include "layer_chassis_dispatch.h"
#include <array>
#include <atomic>
#include <functional>
#include <list>
#include <map>
#include <memory>
#include <set>
#include <string.h>
#include <unordered_map>
#include <unordered_set>
#include <vector>
#include <memory>
#include <list>
#ifdef VK_USE_PLATFORM_ANDROID_KHR
#include "android_ndk_types.h"
#endif // VK_USE_PLATFORM_ANDROID_KHR
// Fwd declarations -- including descriptor_set.h creates an ugly include loop
namespace cvdescriptorset {
class DescriptorSetLayoutDef;
class DescriptorSetLayout;
class DescriptorSet;
} // namespace cvdescriptorset
struct CMD_BUFFER_STATE;
class CoreChecks;
class ValidationStateTracker;
enum CALL_STATE {
UNCALLED, // Function has not been called
QUERY_COUNT, // Function called once to query a count
QUERY_DETAILS, // Function called w/ a count to query details
};
class BASE_NODE {
public:
// Track when object is being used by an in-flight command buffer
std::atomic_int in_use;
// Track command buffers that this object is bound to
// binding initialized when cmd referencing object is bound to command buffer
// binding removed when command buffer is reset or destroyed
// When an object is destroyed, any bound cbs are set to INVALID.
// "int" value is an index into object_bindings where the corresponding
// backpointer to this node is stored.
small_unordered_map<CMD_BUFFER_STATE *, int, 8> cb_bindings;
// Set to true when the API-level object is destroyed, but this object may
// hang around until its shared_ptr refcount goes to zero.
bool destroyed;
BASE_NODE() {
in_use.store(0);
destroyed = false;
};
};
// Track command pools and their command buffers
struct COMMAND_POOL_STATE : public BASE_NODE {
VkCommandPoolCreateFlags createFlags;
uint32_t queueFamilyIndex;
// Cmd buffers allocated from this pool
std::unordered_set<VkCommandBuffer> commandBuffers;
};
// Utilities for barriers and the commmand pool
template <typename Barrier>
static bool IsTransferOp(const Barrier *barrier) {
return barrier->srcQueueFamilyIndex != barrier->dstQueueFamilyIndex;
}
template <typename Barrier, bool assume_transfer = false>
static bool TempIsReleaseOp(const COMMAND_POOL_STATE *pool, const Barrier *barrier) {
return (assume_transfer || IsTransferOp(barrier)) && (pool->queueFamilyIndex == barrier->srcQueueFamilyIndex);
}
template <typename Barrier, bool assume_transfer = false>
static bool IsAcquireOp(const COMMAND_POOL_STATE *pool, const Barrier *barrier) {
return (assume_transfer || IsTransferOp(barrier)) && (pool->queueFamilyIndex == barrier->dstQueueFamilyIndex);
}
static inline bool QueueFamilyIsSpecial(const uint32_t queue_family_index) {
return (queue_family_index == VK_QUEUE_FAMILY_EXTERNAL_KHR) || (queue_family_index == VK_QUEUE_FAMILY_FOREIGN_EXT);
}
static inline bool QueueFamilyIsIgnored(uint32_t queue_family_index) { return queue_family_index == VK_QUEUE_FAMILY_IGNORED; }
// Intentionally ignore VulkanTypedHandle::node, it is optional
inline bool operator==(const VulkanTypedHandle &a, const VulkanTypedHandle &b) NOEXCEPT {
return a.handle == b.handle && a.type == b.type;
}
namespace std {
template <>
struct hash<VulkanTypedHandle> {
size_t operator()(VulkanTypedHandle obj) const NOEXCEPT { return hash<uint64_t>()(obj.handle) ^ hash<uint32_t>()(obj.type); }
};
} // namespace std
// Flags describing requirements imposed by the pipeline on a descriptor. These
// can't be checked at pipeline creation time as they depend on the Image or
// ImageView bound.
enum descriptor_req {
DESCRIPTOR_REQ_VIEW_TYPE_1D = 1 << VK_IMAGE_VIEW_TYPE_1D,
DESCRIPTOR_REQ_VIEW_TYPE_1D_ARRAY = 1 << VK_IMAGE_VIEW_TYPE_1D_ARRAY,
DESCRIPTOR_REQ_VIEW_TYPE_2D = 1 << VK_IMAGE_VIEW_TYPE_2D,
DESCRIPTOR_REQ_VIEW_TYPE_2D_ARRAY = 1 << VK_IMAGE_VIEW_TYPE_2D_ARRAY,
DESCRIPTOR_REQ_VIEW_TYPE_3D = 1 << VK_IMAGE_VIEW_TYPE_3D,
DESCRIPTOR_REQ_VIEW_TYPE_CUBE = 1 << VK_IMAGE_VIEW_TYPE_CUBE,
DESCRIPTOR_REQ_VIEW_TYPE_CUBE_ARRAY = 1 << VK_IMAGE_VIEW_TYPE_CUBE_ARRAY,
DESCRIPTOR_REQ_ALL_VIEW_TYPE_BITS = (1 << (VK_IMAGE_VIEW_TYPE_END_RANGE + 1)) - 1,
DESCRIPTOR_REQ_SINGLE_SAMPLE = 2 << VK_IMAGE_VIEW_TYPE_END_RANGE,
DESCRIPTOR_REQ_MULTI_SAMPLE = DESCRIPTOR_REQ_SINGLE_SAMPLE << 1,
DESCRIPTOR_REQ_COMPONENT_TYPE_FLOAT = DESCRIPTOR_REQ_MULTI_SAMPLE << 1,
DESCRIPTOR_REQ_COMPONENT_TYPE_SINT = DESCRIPTOR_REQ_COMPONENT_TYPE_FLOAT << 1,
DESCRIPTOR_REQ_COMPONENT_TYPE_UINT = DESCRIPTOR_REQ_COMPONENT_TYPE_SINT << 1,
};
extern unsigned DescriptorRequirementsBitsFromFormat(VkFormat fmt);
typedef std::map<uint32_t, descriptor_req> BindingReqMap;
struct DESCRIPTOR_POOL_STATE : BASE_NODE {
VkDescriptorPool pool;
uint32_t maxSets; // Max descriptor sets allowed in this pool
uint32_t availableSets; // Available descriptor sets in this pool
safe_VkDescriptorPoolCreateInfo createInfo;
std::unordered_set<cvdescriptorset::DescriptorSet *> sets; // Collection of all sets in this pool
std::map<uint32_t, uint32_t> maxDescriptorTypeCount; // Max # of descriptors of each type in this pool
std::map<uint32_t, uint32_t> availableDescriptorTypeCount; // Available # of descriptors of each type in this pool
DESCRIPTOR_POOL_STATE(const VkDescriptorPool pool, const VkDescriptorPoolCreateInfo *pCreateInfo)
: pool(pool),
maxSets(pCreateInfo->maxSets),
availableSets(pCreateInfo->maxSets),
createInfo(pCreateInfo),
maxDescriptorTypeCount(),
availableDescriptorTypeCount() {
// Collect maximums per descriptor type.
for (uint32_t i = 0; i < createInfo.poolSizeCount; ++i) {
uint32_t typeIndex = static_cast<uint32_t>(createInfo.pPoolSizes[i].type);
// Same descriptor types can appear several times
maxDescriptorTypeCount[typeIndex] += createInfo.pPoolSizes[i].descriptorCount;
availableDescriptorTypeCount[typeIndex] = maxDescriptorTypeCount[typeIndex];
}
}
};
// Generic memory binding struct to track objects bound to objects
struct MEM_BINDING {
VkDeviceMemory mem;
VkDeviceSize offset;
VkDeviceSize size;
};
struct BufferBinding {
VkBuffer buffer;
VkDeviceSize size;
VkDeviceSize offset;
};
struct IndexBufferBinding : BufferBinding {
VkIndexType index_type;
};
inline bool operator==(MEM_BINDING a, MEM_BINDING b) NOEXCEPT { return a.mem == b.mem && a.offset == b.offset && a.size == b.size; }
namespace std {
template <>
struct hash<MEM_BINDING> {
size_t operator()(MEM_BINDING mb) const NOEXCEPT {
auto intermediate = hash<uint64_t>()(reinterpret_cast<uint64_t &>(mb.mem)) ^ hash<uint64_t>()(mb.offset);
return intermediate ^ hash<uint64_t>()(mb.size);
}
};
} // namespace std
// Superclass for bindable object state (currently images and buffers)
class BINDABLE : public BASE_NODE {
public:
bool sparse; // Is this object being bound with sparse memory or not?
// Non-sparse binding data
MEM_BINDING binding;
// Memory requirements for this BINDABLE
VkMemoryRequirements requirements;
// bool to track if memory requirements were checked
bool memory_requirements_checked;
// Tracks external memory types creating resource
VkExternalMemoryHandleTypeFlags external_memory_handle;
// Sparse binding data, initially just tracking MEM_BINDING per mem object
// There's more data for sparse bindings so need better long-term solution
// TODO : Need to update solution to track all sparse binding data
std::unordered_set<MEM_BINDING> sparse_bindings;
small_unordered_set<VkDeviceMemory, 1> bound_memory_set_;
BINDABLE()
: sparse(false),
binding{},
requirements{},
memory_requirements_checked(false),
external_memory_handle(0),
sparse_bindings{},
bound_memory_set_{} {};
// Update the cached set of memory bindings.
// Code that changes binding.mem or sparse_bindings must call UpdateBoundMemorySet()
void UpdateBoundMemorySet() {
bound_memory_set_.clear();
if (!sparse) {
bound_memory_set_.insert(binding.mem);
} else {
for (auto sb : sparse_bindings) {
bound_memory_set_.insert(sb.mem);
}
}
}
// Return unordered set of memory objects that are bound
// Instead of creating a set from scratch each query, return the cached one
const small_unordered_set<VkDeviceMemory, 1> &GetBoundMemory() const { return bound_memory_set_; }
};
class BUFFER_STATE : public BINDABLE {
public:
VkBuffer buffer;
VkBufferCreateInfo createInfo;
VkDeviceAddress deviceAddress;
BUFFER_STATE(VkBuffer buff, const VkBufferCreateInfo *pCreateInfo) : buffer(buff), createInfo(*pCreateInfo) {
if ((createInfo.sharingMode == VK_SHARING_MODE_CONCURRENT) && (createInfo.queueFamilyIndexCount > 0)) {
uint32_t *pQueueFamilyIndices = new uint32_t[createInfo.queueFamilyIndexCount];
for (uint32_t i = 0; i < createInfo.queueFamilyIndexCount; i++) {
pQueueFamilyIndices[i] = pCreateInfo->pQueueFamilyIndices[i];
}
createInfo.pQueueFamilyIndices = pQueueFamilyIndices;
}
if (createInfo.flags & VK_BUFFER_CREATE_SPARSE_BINDING_BIT) {
sparse = true;
}
auto *externalMemoryInfo = lvl_find_in_chain<VkExternalMemoryBufferCreateInfo>(pCreateInfo->pNext);
if (externalMemoryInfo) {
external_memory_handle = externalMemoryInfo->handleTypes;
}
};
BUFFER_STATE(BUFFER_STATE const &rh_obj) = delete;
~BUFFER_STATE() {
if ((createInfo.sharingMode == VK_SHARING_MODE_CONCURRENT) && (createInfo.queueFamilyIndexCount > 0)) {
delete[] createInfo.pQueueFamilyIndices;
createInfo.pQueueFamilyIndices = nullptr;
}
};
};
class BUFFER_VIEW_STATE : public BASE_NODE {
public:
VkBufferView buffer_view;
VkBufferViewCreateInfo create_info;
std::shared_ptr<BUFFER_STATE> buffer_state;
BUFFER_VIEW_STATE(const std::shared_ptr<BUFFER_STATE> &bf, VkBufferView bv, const VkBufferViewCreateInfo *ci)
: buffer_view(bv), create_info(*ci), buffer_state(bf){};
BUFFER_VIEW_STATE(const BUFFER_VIEW_STATE &rh_obj) = delete;
};
struct SAMPLER_STATE : public BASE_NODE {
VkSampler sampler;
VkSamplerCreateInfo createInfo;
VkSamplerYcbcrConversion samplerConversion = VK_NULL_HANDLE;
SAMPLER_STATE(const VkSampler *ps, const VkSamplerCreateInfo *pci) : sampler(*ps), createInfo(*pci) {
auto *conversionInfo = lvl_find_in_chain<VkSamplerYcbcrConversionInfo>(pci->pNext);
if (conversionInfo) samplerConversion = conversionInfo->conversion;
}
};
class IMAGE_STATE : public BINDABLE {
public:
VkImage image;
VkImageCreateInfo createInfo;
bool valid; // If this is a swapchain image backing memory track valid here as it doesn't have DEVICE_MEMORY_STATE
bool acquired; // If this is a swapchain image, has it been acquired by the app.
bool shared_presentable; // True for a front-buffered swapchain image
bool layout_locked; // A front-buffered image that has been presented can never have layout transitioned
bool get_sparse_reqs_called; // Track if GetImageSparseMemoryRequirements() has been called for this image
bool sparse_metadata_required; // Track if sparse metadata aspect is required for this image
bool sparse_metadata_bound; // Track if sparse metadata aspect is bound to this image
bool imported_ahb; // True if image was imported from an Android Hardware Buffer
bool has_ahb_format; // True if image was created with an external Android format
uint64_t ahb_format; // External Android format, if provided
VkImageSubresourceRange full_range; // The normalized ISR for all levels, layers (slices), and aspects
VkSwapchainKHR create_from_swapchain;
VkSwapchainKHR bind_swapchain;
uint32_t bind_swapchain_imageIndex;
#ifdef VK_USE_PLATFORM_ANDROID_KHR
uint64_t external_format_android;
#endif // VK_USE_PLATFORM_ANDROID_KHR
std::vector<VkSparseImageMemoryRequirements> sparse_requirements;
IMAGE_STATE(VkImage img, const VkImageCreateInfo *pCreateInfo);
IMAGE_STATE(IMAGE_STATE const &rh_obj) = delete;
std::unordered_set<VkImage> aliasing_images;
bool IsCompatibleAliasing(IMAGE_STATE *other_image_state);
bool IsCreateInfoEqual(const VkImageCreateInfo &other_createInfo) const;
bool IsCreateInfoDedicatedAllocationImageAliasingCompatible(const VkImageCreateInfo &other_createInfo) const;
inline bool IsImageTypeEqual(const VkImageCreateInfo &other_createInfo) const {
return createInfo.imageType == other_createInfo.imageType;
}
inline bool IsFormatEqual(const VkImageCreateInfo &other_createInfo) const {
return createInfo.format == other_createInfo.format;
}
inline bool IsMipLevelsEqual(const VkImageCreateInfo &other_createInfo) const {
return createInfo.mipLevels == other_createInfo.mipLevels;
}
inline bool IsUsageEqual(const VkImageCreateInfo &other_createInfo) const { return createInfo.usage == other_createInfo.usage; }
inline bool IsSamplesEqual(const VkImageCreateInfo &other_createInfo) const {
return createInfo.samples == other_createInfo.samples;
}
inline bool IsTilingEqual(const VkImageCreateInfo &other_createInfo) const {
return createInfo.tiling == other_createInfo.tiling;
}
inline bool IsArrayLayersEqual(const VkImageCreateInfo &other_createInfo) const {
return createInfo.arrayLayers == other_createInfo.arrayLayers;
}
inline bool IsInitialLayoutEqual(const VkImageCreateInfo &other_createInfo) const {
return createInfo.initialLayout == other_createInfo.initialLayout;
}
inline bool IsSharingModeEqual(const VkImageCreateInfo &other_createInfo) const {
return createInfo.sharingMode == other_createInfo.sharingMode;
}
inline bool IsExtentEqual(const VkImageCreateInfo &other_createInfo) const {
return (createInfo.extent.width == other_createInfo.extent.width) &&
(createInfo.extent.height == other_createInfo.extent.height) &&
(createInfo.extent.depth == other_createInfo.extent.depth);
}
inline bool IsQueueFamilyIndicesEqual(const VkImageCreateInfo &other_createInfo) const {
return (createInfo.queueFamilyIndexCount == other_createInfo.queueFamilyIndexCount) &&
(createInfo.queueFamilyIndexCount == 0 ||
memcmp(createInfo.pQueueFamilyIndices, other_createInfo.pQueueFamilyIndices,
createInfo.queueFamilyIndexCount * sizeof(createInfo.pQueueFamilyIndices[0])) == 0);
}
~IMAGE_STATE() {
if ((createInfo.sharingMode == VK_SHARING_MODE_CONCURRENT) && (createInfo.queueFamilyIndexCount > 0)) {
delete[] createInfo.pQueueFamilyIndices;
createInfo.pQueueFamilyIndices = nullptr;
}
};
};
class IMAGE_VIEW_STATE : public BASE_NODE {
public:
VkImageView image_view;
VkImageViewCreateInfo create_info;
VkImageSubresourceRange normalized_subresource_range;
VkSampleCountFlagBits samples;
unsigned descriptor_format_bits;
VkSamplerYcbcrConversion samplerConversion; // Handle of the ycbcr sampler conversion the image was created with, if any
std::shared_ptr<IMAGE_STATE> image_state;
IMAGE_VIEW_STATE(const std::shared_ptr<IMAGE_STATE> &image_state, VkImageView iv, const VkImageViewCreateInfo *ci);
IMAGE_VIEW_STATE(const IMAGE_VIEW_STATE &rh_obj) = delete;
};
class ACCELERATION_STRUCTURE_STATE : public BINDABLE {
public:
VkAccelerationStructureNV acceleration_structure;
safe_VkAccelerationStructureCreateInfoNV create_info;
bool memory_requirements_checked = false;
VkMemoryRequirements2KHR memory_requirements;
bool build_scratch_memory_requirements_checked = false;
VkMemoryRequirements2KHR build_scratch_memory_requirements;
bool update_scratch_memory_requirements_checked = false;
VkMemoryRequirements2KHR update_scratch_memory_requirements;
bool built = false;
safe_VkAccelerationStructureInfoNV build_info;
uint64_t opaque_handle = 0;
ACCELERATION_STRUCTURE_STATE(VkAccelerationStructureNV as, const VkAccelerationStructureCreateInfoNV *ci)
: acceleration_structure(as),
create_info(ci),
memory_requirements{},
build_scratch_memory_requirements_checked{},
update_scratch_memory_requirements_checked{} {}
ACCELERATION_STRUCTURE_STATE(const ACCELERATION_STRUCTURE_STATE &rh_obj) = delete;
};
struct MemRange {
VkDeviceSize offset = 0;
VkDeviceSize size = 0;
};
// Data struct for tracking memory object
struct DEVICE_MEMORY_STATE : public BASE_NODE {
void *object; // Dispatchable object used to create this memory (device of swapchain)
VkDeviceMemory mem;
safe_VkMemoryAllocateInfo alloc_info;
bool is_dedicated;
VkBuffer dedicated_buffer;
VkImage dedicated_image;
bool is_export;
VkExternalMemoryHandleTypeFlags export_handle_type_flags;
std::unordered_set<VulkanTypedHandle> obj_bindings; // objects bound to this memory
// Convenience vectors of handles to speed up iterating over objects independently
std::unordered_set<VkImage> bound_images;
// TODO: Convert the two sets to the correct types.
std::unordered_set<uint64_t> bound_buffers;
std::unordered_set<uint64_t> bound_acceleration_structures;
MemRange mapped_range;
void *shadow_copy_base; // Base of layer's allocation for guard band, data, and alignment space
void *shadow_copy; // Pointer to start of guard-band data before mapped region
uint64_t shadow_pad_size; // Size of the guard-band data before and after actual data. It MUST be a
// multiple of limits.minMemoryMapAlignment
void *p_driver_data; // Pointer to application's actual memory
DEVICE_MEMORY_STATE(void *disp_object, const VkDeviceMemory in_mem, const VkMemoryAllocateInfo *p_alloc_info)
: object(disp_object),
mem(in_mem),
alloc_info(p_alloc_info),
is_dedicated(false),
dedicated_buffer(VK_NULL_HANDLE),
dedicated_image(VK_NULL_HANDLE),
is_export(false),
export_handle_type_flags(0),
mapped_range{},
shadow_copy_base(0),
shadow_copy(0),
shadow_pad_size(0),
p_driver_data(0){};
};
struct SWAPCHAIN_IMAGE {
VkImage image;
std::unordered_set<VkImage> bound_images;
};
class SWAPCHAIN_NODE : public BASE_NODE {
public:
safe_VkSwapchainCreateInfoKHR createInfo;
VkSwapchainKHR swapchain;
std::vector<SWAPCHAIN_IMAGE> images;
bool retired = false;
bool shared_presentable = false;
CALL_STATE vkGetSwapchainImagesKHRState = UNCALLED;
uint32_t get_swapchain_image_count = 0;
SWAPCHAIN_NODE(const VkSwapchainCreateInfoKHR *pCreateInfo, VkSwapchainKHR swapchain)
: createInfo(pCreateInfo), swapchain(swapchain) {}
};
struct ColorAspectTraits {
static const uint32_t kAspectCount = 1;
static int Index(VkImageAspectFlags mask) { return 0; };
static VkImageAspectFlags AspectMask() { return VK_IMAGE_ASPECT_COLOR_BIT; }
static const std::array<VkImageAspectFlagBits, kAspectCount> &AspectBits() {
static std::array<VkImageAspectFlagBits, kAspectCount> kAspectBits{{VK_IMAGE_ASPECT_COLOR_BIT}};
return kAspectBits;
}
};
struct DepthAspectTraits {
static const uint32_t kAspectCount = 1;
static int Index(VkImageAspectFlags mask) { return 0; };
static VkImageAspectFlags AspectMask() { return VK_IMAGE_ASPECT_DEPTH_BIT; }
static const std::array<VkImageAspectFlagBits, kAspectCount> &AspectBits() {
static std::array<VkImageAspectFlagBits, kAspectCount> kAspectBits{{VK_IMAGE_ASPECT_DEPTH_BIT}};
return kAspectBits;
}
};
struct StencilAspectTraits {
static const uint32_t kAspectCount = 1;
static int Index(VkImageAspectFlags mask) { return 0; };
static VkImageAspectFlags AspectMask() { return VK_IMAGE_ASPECT_STENCIL_BIT; }
static const std::array<VkImageAspectFlagBits, kAspectCount> &AspectBits() {
static std::array<VkImageAspectFlagBits, kAspectCount> kAspectBits{{VK_IMAGE_ASPECT_STENCIL_BIT}};
return kAspectBits;
}
};
struct DepthStencilAspectTraits {
// VK_IMAGE_ASPECT_DEPTH_BIT = 0x00000002, >> 1 -> 1 -1 -> 0
// VK_IMAGE_ASPECT_STENCIL_BIT = 0x00000004, >> 1 -> 2 -1 = 1
static const uint32_t kAspectCount = 2;
static uint32_t Index(VkImageAspectFlags mask) {
uint32_t index = (mask >> 1) - 1;
assert((index == 0) || (index == 1));
return index;
};
static VkImageAspectFlags AspectMask() { return VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT; }
static const std::array<VkImageAspectFlagBits, kAspectCount> &AspectBits() {
static std::array<VkImageAspectFlagBits, kAspectCount> kAspectBits{
{VK_IMAGE_ASPECT_DEPTH_BIT, VK_IMAGE_ASPECT_STENCIL_BIT}};
return kAspectBits;
}
};
struct Multiplane2AspectTraits {
// VK_IMAGE_ASPECT_PLANE_0_BIT = 0x00000010, >> 4 - 1 -> 0
// VK_IMAGE_ASPECT_PLANE_1_BIT = 0x00000020, >> 4 - 1 -> 1
static const uint32_t kAspectCount = 2;
static uint32_t Index(VkImageAspectFlags mask) {
uint32_t index = (mask >> 4) - 1;
assert((index == 0) || (index == 1));
return index;
};
static VkImageAspectFlags AspectMask() { return VK_IMAGE_ASPECT_PLANE_0_BIT | VK_IMAGE_ASPECT_PLANE_1_BIT; }
static const std::array<VkImageAspectFlagBits, kAspectCount> &AspectBits() {
static std::array<VkImageAspectFlagBits, kAspectCount> kAspectBits{
{VK_IMAGE_ASPECT_PLANE_0_BIT, VK_IMAGE_ASPECT_PLANE_1_BIT}};
return kAspectBits;
}
};
struct Multiplane3AspectTraits {
// VK_IMAGE_ASPECT_PLANE_0_BIT = 0x00000010, >> 4 - 1 -> 0
// VK_IMAGE_ASPECT_PLANE_1_BIT = 0x00000020, >> 4 - 1 -> 1
// VK_IMAGE_ASPECT_PLANE_2_BIT = 0x00000040, >> 4 - 1 -> 3
static const uint32_t kAspectCount = 3;
static uint32_t Index(VkImageAspectFlags mask) {
uint32_t index = (mask >> 4) - 1;
index = index > 2 ? 2 : index;
assert((index == 0) || (index == 1) || (index == 2));
return index;
};
static VkImageAspectFlags AspectMask() {
return VK_IMAGE_ASPECT_PLANE_0_BIT | VK_IMAGE_ASPECT_PLANE_1_BIT | VK_IMAGE_ASPECT_PLANE_2_BIT;
}
static const std::array<VkImageAspectFlagBits, kAspectCount> &AspectBits() {
static std::array<VkImageAspectFlagBits, kAspectCount> kAspectBits{
{VK_IMAGE_ASPECT_PLANE_0_BIT, VK_IMAGE_ASPECT_PLANE_1_BIT, VK_IMAGE_ASPECT_PLANE_2_BIT}};
return kAspectBits;
}
};
std::string FormatDebugLabel(const char *prefix, const LoggingLabel &label);
const static VkImageLayout kInvalidLayout = VK_IMAGE_LAYOUT_MAX_ENUM;
// Interface class.
class ImageSubresourceLayoutMap {
public:
typedef std::function<bool(const VkImageSubresource &, VkImageLayout, VkImageLayout)> Callback;
struct InitialLayoutState {
VkImageView image_view; // For relaxed matching rule evaluation, else VK_NULL_HANDLE
VkImageAspectFlags aspect_mask; // For relaxed matching rules... else 0
LoggingLabel label;
InitialLayoutState(const CMD_BUFFER_STATE &cb_state_, const IMAGE_VIEW_STATE *view_state);
InitialLayoutState() : image_view(VK_NULL_HANDLE), aspect_mask(0), label() {}
};
struct SubresourceLayout {
VkImageSubresource subresource;
VkImageLayout layout;
};
struct SubresourceRangeLayout {
VkImageSubresourceRange range;
VkImageLayout layout;
};
class ConstIteratorInterface {
public:
// Make the value accessor non virtual
const SubresourceLayout &operator*() const { return value_; }
virtual ConstIteratorInterface &operator++() = 0;
virtual bool AtEnd() const = 0;
virtual ~ConstIteratorInterface(){};
protected:
SubresourceLayout value_;
};
class ConstIterator {
public:
ConstIterator &operator++() {
++(*it_);
return *this;
}
const SubresourceLayout &operator*() const { return *(*it_); }
ConstIterator(ConstIteratorInterface *it) : it_(it){};
bool AtEnd() const { return it_->AtEnd(); }
protected:
std::unique_ptr<ConstIteratorInterface> it_;
};
virtual ConstIterator BeginInitialUse() const = 0;
virtual ConstIterator BeginSetLayout() const = 0;
virtual bool SetSubresourceRangeLayout(const CMD_BUFFER_STATE &cb_state, const VkImageSubresourceRange &range,
VkImageLayout layout, VkImageLayout expected_layout = kInvalidLayout) = 0;
virtual bool SetSubresourceRangeInitialLayout(const CMD_BUFFER_STATE &cb_state, const VkImageSubresourceRange &range,
VkImageLayout layout, const IMAGE_VIEW_STATE *view_state = nullptr) = 0;
virtual bool ForRange(const VkImageSubresourceRange &range, const Callback &callback, bool skip_invalid = true,
bool always_get_initial = false) const = 0;
virtual VkImageLayout GetSubresourceLayout(const VkImageSubresource subresource) const = 0;
virtual VkImageLayout GetSubresourceInitialLayout(const VkImageSubresource subresource) const = 0;
virtual const InitialLayoutState *GetSubresourceInitialLayoutState(const VkImageSubresource subresource) const = 0;
virtual bool UpdateFrom(const ImageSubresourceLayoutMap &from) = 0;
virtual uintptr_t CompatibilityKey() const = 0;
ImageSubresourceLayoutMap() {}
virtual ~ImageSubresourceLayoutMap() {}
};
template <typename AspectTraits_, size_t kSparseThreshold = 64U>
class ImageSubresourceLayoutMapImpl : public ImageSubresourceLayoutMap {
public:
typedef ImageSubresourceLayoutMap Base;
typedef AspectTraits_ AspectTraits;
typedef Base::SubresourceLayout SubresourceLayout;
typedef sparse_container::SparseVector<size_t, VkImageLayout, true, kInvalidLayout, kSparseThreshold> LayoutMap;
typedef sparse_container::SparseVector<size_t, VkImageLayout, false, kInvalidLayout, kSparseThreshold> InitialLayoutMap;
struct Layouts {
LayoutMap current;
InitialLayoutMap initial;
Layouts(size_t size) : current(0, size), initial(0, size) {}
};
template <typename Container>
class ConstIteratorImpl : public Base::ConstIteratorInterface {
public:
ConstIteratorImpl &operator++() override {
++it_;
UpdateValue();
return *this;
}
// Just good enough for cend checks
ConstIteratorImpl(const ImageSubresourceLayoutMapImpl &map, const Container &container)
: map_(&map), container_(&container), the_end_(false) {
it_ = container_->cbegin();
UpdateValue();
}
~ConstIteratorImpl() override {}
virtual bool AtEnd() const override { return the_end_; }
protected:
void UpdateValue() {
if (it_ != container_->cend()) {
value_.subresource = map_->Decode((*it_).first);
value_.layout = (*it_).second;
} else {
the_end_ = true;
value_.layout = kInvalidLayout;
}
}
typedef typename Container::const_iterator ContainerIterator;
const ImageSubresourceLayoutMapImpl *map_;
const Container *container_;
bool the_end_;
ContainerIterator it_;
};
Base::ConstIterator BeginInitialUse() const override {
return Base::ConstIterator(new ConstIteratorImpl<InitialLayoutMap>(*this, layouts_.initial));
}
Base::ConstIterator BeginSetLayout() const override {
return Base::ConstIterator(new ConstIteratorImpl<LayoutMap>(*this, layouts_.current));
}
bool SetSubresourceRangeLayout(const CMD_BUFFER_STATE &cb_state, const VkImageSubresourceRange &range, VkImageLayout layout,
VkImageLayout expected_layout = kInvalidLayout) override {
bool updated = false;
if (expected_layout == kInvalidLayout) {
// Set the initial layout to the set layout as we had no other layout to reference
expected_layout = layout;
}
if (!InRange(range)) return false; // Don't even try to track bogus subreources
InitialLayoutState *initial_state = nullptr;
const uint32_t end_mip = range.baseMipLevel + range.levelCount;
const auto &aspects = AspectTraits::AspectBits();
for (uint32_t aspect_index = 0; aspect_index < AspectTraits::kAspectCount; aspect_index++) {
if (0 == (range.aspectMask & aspects[aspect_index])) continue;
size_t array_offset = Encode(aspect_index, range.baseMipLevel);
for (uint32_t mip_level = range.baseMipLevel; mip_level < end_mip; ++mip_level, array_offset += mip_size_) {
size_t start = array_offset + range.baseArrayLayer;
size_t end = start + range.layerCount;
bool updated_level = layouts_.current.SetRange(start, end, layout);
if (updated_level) {
// We only need to try setting the initial layout, if we changed any of the layout values above
updated = true;
if (layouts_.initial.SetRange(start, end, expected_layout)) {
// We only need to try setting the initial layout *state* if the initial layout was updated
initial_state = UpdateInitialLayoutState(start, end, initial_state, cb_state, nullptr);
}
}
}
}
if (updated) version_++;
return updated;
}
bool SetSubresourceRangeInitialLayout(const CMD_BUFFER_STATE &cb_state, const VkImageSubresourceRange &range,
VkImageLayout layout, const IMAGE_VIEW_STATE *view_state = nullptr) override {
bool updated = false;
if (!InRange(range)) return false; // Don't even try to track bogus subreources
InitialLayoutState *initial_state = nullptr;
const uint32_t end_mip = range.baseMipLevel + range.levelCount;
const auto &aspects = AspectTraits::AspectBits();
for (uint32_t aspect_index = 0; aspect_index < AspectTraits::kAspectCount; aspect_index++) {
if (0 == (range.aspectMask & aspects[aspect_index])) continue;
size_t array_offset = Encode(aspect_index, range.baseMipLevel);
for (uint32_t mip_level = range.baseMipLevel; mip_level < end_mip; ++mip_level, array_offset += mip_size_) {
size_t start = array_offset + range.baseArrayLayer;
size_t end = start + range.layerCount;
bool updated_level = layouts_.initial.SetRange(start, end, layout);
if (updated_level) {
updated = true;
// We only need to try setting the initial layout *state* if the initial layout was updated
initial_state = UpdateInitialLayoutState(start, end, initial_state, cb_state, view_state);
}
}
}
if (updated) version_++;
return updated;
}
// Loop over the given range calling the callback, primarily for
// validation checks. By default the initial_value is only looked
// up if the set value isn't found.
bool ForRange(const VkImageSubresourceRange &range, const Callback &callback, bool skip_invalid = true,
bool always_get_initial = false) const override {
if (!InRange(range)) return false; // Don't even try to process bogus subreources
VkImageSubresource subres;
auto &level = subres.mipLevel;
auto &layer = subres.arrayLayer;
auto &aspect = subres.aspectMask;
const auto &aspects = AspectTraits::AspectBits();
bool keep_on = true;
const uint32_t end_mip = range.baseMipLevel + range.levelCount;
const uint32_t end_layer = range.baseArrayLayer + range.layerCount;
for (uint32_t aspect_index = 0; aspect_index < AspectTraits::kAspectCount; aspect_index++) {
if (0 == (range.aspectMask & aspects[aspect_index])) continue;
aspect = aspects[aspect_index]; // noting that this and the following loop indices are references
size_t array_offset = Encode(aspect_index, range.baseMipLevel);
for (level = range.baseMipLevel; level < end_mip; ++level, array_offset += mip_size_) {
for (layer = range.baseArrayLayer; layer < end_layer; layer++) {
// TODO -- would an interator with range check be faster?
size_t index = array_offset + layer;
VkImageLayout layout = layouts_.current.Get(index);
VkImageLayout initial_layout = kInvalidLayout;
if (always_get_initial || (layout == kInvalidLayout)) {
initial_layout = layouts_.initial.Get(index);
}
if (!skip_invalid || (layout != kInvalidLayout) || (initial_layout != kInvalidLayout)) {
keep_on = callback(subres, layout, initial_layout);
if (!keep_on) return keep_on; // False value from the callback aborts the range traversal
}
}
}
}
return keep_on;
}
VkImageLayout GetSubresourceInitialLayout(const VkImageSubresource subresource) const override {
if (!InRange(subresource)) return kInvalidLayout;
uint32_t aspect_index = AspectTraits::Index(subresource.aspectMask);
size_t index = Encode(aspect_index, subresource.mipLevel, subresource.arrayLayer);
return layouts_.initial.Get(index);
}
const InitialLayoutState *GetSubresourceInitialLayoutState(const VkImageSubresource subresource) const override {
if (!InRange(subresource)) return nullptr;
uint32_t aspect_index = AspectTraits::Index(subresource.aspectMask);
size_t index = Encode(aspect_index, subresource.mipLevel, subresource.arrayLayer);
return initial_layout_state_map_.Get(index);
}
VkImageLayout GetSubresourceLayout(const VkImageSubresource subresource) const override {
if (!InRange(subresource)) return kInvalidLayout;
uint32_t aspect_index = AspectTraits::Index(subresource.aspectMask);
size_t index = Encode(aspect_index, subresource.mipLevel, subresource.arrayLayer);
return layouts_.current.Get(index);
}
// TODO: make sure this paranoia check is sufficient and not too much.
uintptr_t CompatibilityKey() const override {
return (reinterpret_cast<const uintptr_t>(&image_state_) ^ AspectTraits::AspectMask() ^ kSparseThreshold);
}
bool UpdateFrom(const ImageSubresourceLayoutMap &other) override {
// Must be from matching images for the reinterpret cast to be valid
assert(CompatibilityKey() == other.CompatibilityKey());
if (CompatibilityKey() != other.CompatibilityKey()) return false;
const auto &from = reinterpret_cast<const ImageSubresourceLayoutMapImpl &>(other);
bool updated = false;
updated |= layouts_.initial.Merge(from.layouts_.initial);
updated |= layouts_.current.Merge(from.layouts_.current);
initial_layout_state_map_.Merge(from.initial_layout_state_map_);
return updated;
}
ImageSubresourceLayoutMapImpl() : Base() {}
ImageSubresourceLayoutMapImpl(const IMAGE_STATE &image_state)
: Base(),
image_state_(image_state),
mip_size_(image_state.full_range.layerCount),
aspect_size_(mip_size_ * image_state.full_range.levelCount),
version_(0),
layouts_(aspect_size_ * AspectTraits::kAspectCount),
initial_layout_states_(),
initial_layout_state_map_(0, aspect_size_ * AspectTraits::kAspectCount) {
// Setup the row <-> aspect/mip_level base Encode/Decode LUT...
aspect_offsets_[0] = 0;
for (size_t i = 1; i < aspect_offsets_.size(); ++i) { // Size is a compile time constant
aspect_offsets_[i] = aspect_offsets_[i - 1] + aspect_size_;
}
}
~ImageSubresourceLayoutMapImpl() override {}
protected:
// This looks a bit ponderous but kAspectCount is a compile time constant
VkImageSubresource Decode(size_t index) const {
VkImageSubresource subres;
// find aspect index
uint32_t aspect_index = 0;
if (AspectTraits::kAspectCount == 2) {
if (index >= aspect_offsets_[1]) {
aspect_index = 1;
index = index - aspect_offsets_[aspect_index];
}
} else if (AspectTraits::kAspectCount == 3) {
if (index >= aspect_offsets_[2]) {
aspect_index = 2;
} else if (index >= aspect_offsets_[1]) {
aspect_index = 1;
}
index = index - aspect_offsets_[aspect_index];
} else {
assert(AspectTraits::kAspectCount == 1); // Only aspect counts of 1, 2, and 3 supported
}
subres.aspectMask = AspectTraits::AspectBits()[aspect_index];
subres.mipLevel =
static_cast<uint32_t>(index / mip_size_); // One hopes the compiler with optimize this pair of divisions...
subres.arrayLayer = static_cast<uint32_t>(index % mip_size_);
return subres;
}
uint32_t LevelLimit(uint32_t level) const { return (std::min)(image_state_.full_range.levelCount, level); }
uint32_t LayerLimit(uint32_t layer) const { return (std::min)(image_state_.full_range.layerCount, layer); }
bool InRange(const VkImageSubresource &subres) const {
bool in_range = (subres.mipLevel < image_state_.full_range.levelCount) &&
(subres.arrayLayer < image_state_.full_range.layerCount) &&
(subres.aspectMask & AspectTraits::AspectMask());
return in_range;
}
bool InRange(const VkImageSubresourceRange &range) const {
bool in_range = (range.baseMipLevel < image_state_.full_range.levelCount) &&
((range.baseMipLevel + range.levelCount) <= image_state_.full_range.levelCount) &&
(range.baseArrayLayer < image_state_.full_range.layerCount) &&
((range.baseArrayLayer + range.layerCount) <= image_state_.full_range.layerCount) &&
(range.aspectMask & AspectTraits::AspectMask());
return in_range;
}
inline size_t Encode(uint32_t aspect_index) const {
return (AspectTraits::kAspectCount == 1) ? 0 : aspect_offsets_[aspect_index];
}
inline size_t Encode(uint32_t aspect_index, uint32_t mip_level) const { return Encode(aspect_index) + mip_level * mip_size_; }
inline size_t Encode(uint32_t aspect_index, uint32_t mip_level, uint32_t array_layer) const {
return Encode(aspect_index, mip_level) + array_layer;
}
InitialLayoutState *UpdateInitialLayoutState(size_t start, size_t end, InitialLayoutState *initial_state,
const CMD_BUFFER_STATE &cb_state, const IMAGE_VIEW_STATE *view_state) {
if (!initial_state) {
// Allocate on demand... initial_layout_states_ holds ownership as a unique_ptr, while
// each subresource has a non-owning copy of the plain pointer.
initial_state = new InitialLayoutState(cb_state, view_state);
initial_layout_states_.emplace_back(initial_state);
}
assert(initial_state);
initial_layout_state_map_.SetRange(start, end, initial_state);
return initial_state;
}
typedef std::vector<std::unique_ptr<InitialLayoutState>> InitialLayoutStates;
// This map *also* needs "write once" semantics
typedef sparse_container::SparseVector<size_t, InitialLayoutState *, false, nullptr, kSparseThreshold> InitialLayoutStateMap;
const IMAGE_STATE &image_state_;
const size_t mip_size_;
const size_t aspect_size_;
uint64_t version_ = 0;
Layouts layouts_;
InitialLayoutStates initial_layout_states_;
InitialLayoutStateMap initial_layout_state_map_;
std::array<size_t, AspectTraits::kAspectCount> aspect_offsets_;
};
static VkImageLayout NormalizeImageLayout(VkImageLayout layout, VkImageLayout non_normal, VkImageLayout normal) {
return (layout == non_normal) ? normal : layout;
}
static VkImageLayout NormalizeDepthImageLayout(VkImageLayout layout) {
return NormalizeImageLayout(layout, VK_IMAGE_LAYOUT_DEPTH_STENCIL_READ_ONLY_OPTIMAL,
VK_IMAGE_LAYOUT_DEPTH_READ_ONLY_STENCIL_ATTACHMENT_OPTIMAL);
}
static VkImageLayout NormalizeStencilImageLayout(VkImageLayout layout) {
return NormalizeImageLayout(layout, VK_IMAGE_LAYOUT_DEPTH_STENCIL_READ_ONLY_OPTIMAL,
VK_IMAGE_LAYOUT_DEPTH_ATTACHMENT_STENCIL_READ_ONLY_OPTIMAL);
}
static bool ImageLayoutMatches(const VkImageAspectFlags aspect_mask, VkImageLayout a, VkImageLayout b) {
bool matches = (a == b);
if (!matches) {
// Relaxed rules when referencing *only* the depth or stencil aspects
if (aspect_mask == VK_IMAGE_ASPECT_DEPTH_BIT) {
matches = NormalizeDepthImageLayout(a) == NormalizeDepthImageLayout(b);
} else if (aspect_mask == VK_IMAGE_ASPECT_STENCIL_BIT) {
matches = NormalizeStencilImageLayout(a) == NormalizeStencilImageLayout(b);
}
}
return matches;
}
// Utility type for ForRange callbacks
struct LayoutUseCheckAndMessage {
const static VkImageAspectFlags kDepthOrStencil = VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT;
const ImageSubresourceLayoutMap *layout_map;
const VkImageAspectFlags aspect_mask;
const char *message;
VkImageLayout layout;
LayoutUseCheckAndMessage() = delete;
LayoutUseCheckAndMessage(const ImageSubresourceLayoutMap *layout_map_, const VkImageAspectFlags aspect_mask_ = 0)
: layout_map(layout_map_), aspect_mask{aspect_mask_}, message(nullptr), layout(kInvalidLayout) {}
bool Check(const VkImageSubresource &subres, VkImageLayout check, VkImageLayout current_layout, VkImageLayout initial_layout) {
message = nullptr;
layout = kInvalidLayout; // Success status
if (current_layout != kInvalidLayout && !ImageLayoutMatches(aspect_mask, check, current_layout)) {
message = "previous known";
layout = current_layout;
} else if ((initial_layout != kInvalidLayout) && !ImageLayoutMatches(aspect_mask, check, initial_layout)) {
// To check the relaxed rule matching we need to see how the initial use was used
const auto initial_layout_state = layout_map->GetSubresourceInitialLayoutState(subres);
assert(initial_layout_state); // If we have an initial layout, we better have a state for it
if (!((initial_layout_state->aspect_mask & kDepthOrStencil) &&
ImageLayoutMatches(initial_layout_state->aspect_mask, check, initial_layout))) {
message = "previously used";
layout = initial_layout;
}
}
return layout == kInvalidLayout;
}
};
// Store the DAG.
struct DAGNode {
uint32_t pass;
std::vector<uint32_t> prev;
std::vector<uint32_t> next;
};
struct RENDER_PASS_STATE : public BASE_NODE {
VkRenderPass renderPass;
safe_VkRenderPassCreateInfo2KHR createInfo;
std::vector<std::vector<uint32_t>> self_dependencies;
std::vector<DAGNode> subpassToNode;
std::unordered_map<uint32_t, bool> attachment_first_read;
RENDER_PASS_STATE(VkRenderPassCreateInfo2KHR const *pCreateInfo) : createInfo(pCreateInfo) {}
RENDER_PASS_STATE(VkRenderPassCreateInfo const *pCreateInfo) {
ConvertVkRenderPassCreateInfoToV2KHR(*pCreateInfo, &createInfo);
}
};
// Autogenerated as part of the vk_validation_error_message.h codegen
enum CMD_TYPE { VUID_CMD_ENUM_LIST(CMD_) };
enum CB_STATE {
CB_NEW, // Newly created CB w/o any cmds
CB_RECORDING, // BeginCB has been called on this CB
CB_RECORDED, // EndCB has been called on this CB
CB_INVALID_COMPLETE, // had a complete recording, but was since invalidated
CB_INVALID_INCOMPLETE, // fouled before recording was completed
};
// CB Status -- used to track status of various bindings on cmd buffer objects
typedef VkFlags CBStatusFlags;
enum CBStatusFlagBits {
// clang-format off
CBSTATUS_NONE = 0x00000000, // No status is set
CBSTATUS_LINE_WIDTH_SET = 0x00000001, // Line width has been set
CBSTATUS_DEPTH_BIAS_SET = 0x00000002, // Depth bias has been set
CBSTATUS_BLEND_CONSTANTS_SET = 0x00000004, // Blend constants state has been set
CBSTATUS_DEPTH_BOUNDS_SET = 0x00000008, // Depth bounds state object has been set
CBSTATUS_STENCIL_READ_MASK_SET = 0x00000010, // Stencil read mask has been set
CBSTATUS_STENCIL_WRITE_MASK_SET = 0x00000020, // Stencil write mask has been set
CBSTATUS_STENCIL_REFERENCE_SET = 0x00000040, // Stencil reference has been set
CBSTATUS_VIEWPORT_SET = 0x00000080,
CBSTATUS_SCISSOR_SET = 0x00000100,
CBSTATUS_INDEX_BUFFER_BOUND = 0x00000200, // Index buffer has been set
CBSTATUS_EXCLUSIVE_SCISSOR_SET = 0x00000400,
CBSTATUS_SHADING_RATE_PALETTE_SET = 0x00000800,
CBSTATUS_LINE_STIPPLE_SET = 0x00001000,
CBSTATUS_VIEWPORT_W_SCALING_SET = 0x00002000,
CBSTATUS_ALL_STATE_SET = 0x00003DFF, // All state set (intentionally exclude index buffer)
// clang-format on
};
struct QueryObject {
VkQueryPool pool;
uint32_t query;
// These next two fields are *not* used in hash or comparison, they are effectively a data payload
uint32_t index; // must be zero if !indexed
bool indexed;
// Command index in the command buffer where the end of the query was
// recorded (equal to the number of commands in the command buffer before
// the end of the query).
uint64_t endCommandIndex;
QueryObject(VkQueryPool pool_, uint32_t query_) : pool(pool_), query(query_), index(0), indexed(false), endCommandIndex(0) {}
QueryObject(VkQueryPool pool_, uint32_t query_, uint32_t index_)
: pool(pool_), query(query_), index(index_), indexed(true), endCommandIndex(0) {}
QueryObject(const QueryObject &obj)
: pool(obj.pool), query(obj.query), index(obj.index), indexed(obj.indexed), endCommandIndex(obj.endCommandIndex) {}
bool operator<(const QueryObject &rhs) const { return (pool == rhs.pool) ? query < rhs.query : pool < rhs.pool; }
};
inline bool operator==(const QueryObject &query1, const QueryObject &query2) {
return ((query1.pool == query2.pool) && (query1.query == query2.query));
}
struct QueryObjectPass {
QueryObject obj;
uint32_t perf_pass;
QueryObjectPass(const QueryObject &obj_, uint32_t perf_pass_) : obj(obj_), perf_pass(perf_pass_) {}
QueryObjectPass(const QueryObjectPass &obj_) : obj(obj_.obj), perf_pass(obj_.perf_pass) {}
bool operator<(const QueryObjectPass &rhs) const { return (obj == rhs.obj) ? perf_pass < rhs.perf_pass : obj < rhs.obj; }
};
inline bool operator==(const QueryObjectPass &query1, const QueryObjectPass &query2) {
return ((query1.obj == query2.obj) && (query1.perf_pass == query2.perf_pass));
}
enum QueryState {
QUERYSTATE_UNKNOWN, // Initial state.
QUERYSTATE_RESET, // After resetting.
QUERYSTATE_RUNNING, // Query running.
QUERYSTATE_ENDED, // Query ended but results may not be available.
QUERYSTATE_AVAILABLE, // Results available.
};
enum QueryResultType {
QUERYRESULT_UNKNOWN,
QUERYRESULT_NO_DATA,
QUERYRESULT_MAYBE_NO_DATA,
QUERYRESULT_SOME_DATA,
QUERYRESULT_WAIT_ON_RESET,
QUERYRESULT_WAIT_ON_RUNNING,
};
inline const char *string_QueryResultType(QueryResultType result_type) {
switch (result_type) {
case QUERYRESULT_UNKNOWN:
return "query may be in an unknown state";
case QUERYRESULT_NO_DATA:
case QUERYRESULT_MAYBE_NO_DATA:
return "query may return no data";
case QUERYRESULT_SOME_DATA:
return "query will return some data or availability bit";
case QUERYRESULT_WAIT_ON_RESET:
return "waiting on a query that has been reset and not issued yet";
case QUERYRESULT_WAIT_ON_RUNNING:
return "waiting on a query that has not ended yet";
}
assert(false);
return "UNKNOWN QUERY STATE"; // Unreachable.
}
namespace std {
template <>
struct hash<QueryObject> {
size_t operator()(QueryObject query) const throw() {
return hash<uint64_t>()((uint64_t)(query.pool)) ^ hash<uint32_t>()(query.query);
}
};
template <>
struct hash<QueryObjectPass> {
size_t operator()(QueryObjectPass query) const throw() {
return hash<QueryObject>()(query.obj) ^ hash<uint32_t>()(query.perf_pass);
}
};
} // namespace std
struct CBVertexBufferBindingInfo {
std::vector<BufferBinding> vertex_buffer_bindings;
};
struct ImageSubresourcePair {
VkImage image;
bool hasSubresource;
VkImageSubresource subresource;
};
inline bool operator==(const ImageSubresourcePair &img1, const ImageSubresourcePair &img2) {
if (img1.image != img2.image || img1.hasSubresource != img2.hasSubresource) return false;
return !img1.hasSubresource ||
(img1.subresource.aspectMask == img2.subresource.aspectMask && img1.subresource.mipLevel == img2.subresource.mipLevel &&
img1.subresource.arrayLayer == img2.subresource.arrayLayer);
}
namespace std {
template <>
struct hash<ImageSubresourcePair> {
size_t operator()(ImageSubresourcePair img) const throw() {
size_t hashVal = hash<uint64_t>()(reinterpret_cast<uint64_t &>(img.image));
hashVal ^= hash<bool>()(img.hasSubresource);
if (img.hasSubresource) {
hashVal ^= hash<uint32_t>()(reinterpret_cast<uint32_t &>(img.subresource.aspectMask));
hashVal ^= hash<uint32_t>()(img.subresource.mipLevel);
hashVal ^= hash<uint32_t>()(img.subresource.arrayLayer);
}
return hashVal;
}
};
} // namespace std
// Canonical dictionary for PushConstantRanges
using PushConstantRangesDict = hash_util::Dictionary<PushConstantRanges>;
using PushConstantRangesId = PushConstantRangesDict::Id;
// Canonical dictionary for the pipeline layout's layout of descriptorsetlayouts
using DescriptorSetLayoutDef = cvdescriptorset::DescriptorSetLayoutDef;
using DescriptorSetLayoutId = std::shared_ptr<const DescriptorSetLayoutDef>;
using PipelineLayoutSetLayoutsDef = std::vector<DescriptorSetLayoutId>;
using PipelineLayoutSetLayoutsDict =
hash_util::Dictionary<PipelineLayoutSetLayoutsDef, hash_util::IsOrderedContainer<PipelineLayoutSetLayoutsDef>>;
using PipelineLayoutSetLayoutsId = PipelineLayoutSetLayoutsDict::Id;
// Defines/stores a compatibility defintion for set N
// The "layout layout" must store at least set+1 entries, but only the first set+1 are considered for hash and equality testing
// Note: the "cannonical" data are referenced by Id, not including handle or device specific state
// Note: hash and equality only consider layout_id entries [0, set] for determining uniqueness
struct PipelineLayoutCompatDef {
uint32_t set;
PushConstantRangesId push_constant_ranges;
PipelineLayoutSetLayoutsId set_layouts_id;
PipelineLayoutCompatDef(const uint32_t set_index, const PushConstantRangesId pcr_id, const PipelineLayoutSetLayoutsId sl_id)
: set(set_index), push_constant_ranges(pcr_id), set_layouts_id(sl_id) {}
size_t hash() const;
bool operator==(const PipelineLayoutCompatDef &other) const;
};
// Canonical dictionary for PipelineLayoutCompat records
using PipelineLayoutCompatDict = hash_util::Dictionary<PipelineLayoutCompatDef, hash_util::HasHashMember<PipelineLayoutCompatDef>>;
using PipelineLayoutCompatId = PipelineLayoutCompatDict::Id;
// Store layouts and pushconstants for PipelineLayout
struct PIPELINE_LAYOUT_STATE : public BASE_NODE {
VkPipelineLayout layout;
std::vector<std::shared_ptr<cvdescriptorset::DescriptorSetLayout const>> set_layouts;
PushConstantRangesId push_constant_ranges;
std::vector<PipelineLayoutCompatId> compat_for_set;
PIPELINE_LAYOUT_STATE() : layout(VK_NULL_HANDLE), set_layouts{}, push_constant_ranges{}, compat_for_set{} {}
void reset() {
layout = VK_NULL_HANDLE;
set_layouts.clear();
push_constant_ranges.reset();
compat_for_set.clear();
}
};
// Shader typedefs needed to store StageStage below
struct interface_var {
uint32_t id;
uint32_t type_id;
uint32_t offset;
bool is_patch;
bool is_block_member;
bool is_relaxed_precision;
// TODO: collect the name, too? Isn't required to be present.
};
typedef std::pair<unsigned, unsigned> descriptor_slot_t;
class PIPELINE_STATE : public BASE_NODE {
public:
struct StageState {
std::unordered_set<uint32_t> accessible_ids;
std::vector<std::pair<descriptor_slot_t, interface_var>> descriptor_uses;
bool has_writable_descriptor;
};
VkPipeline pipeline;
safe_VkGraphicsPipelineCreateInfo graphicsPipelineCI;
safe_VkComputePipelineCreateInfo computePipelineCI;
safe_VkRayTracingPipelineCreateInfoNV raytracingPipelineCI;
// Hold shared ptr to RP in case RP itself is destroyed
std::shared_ptr<const RENDER_PASS_STATE> rp_state;
// Flag of which shader stages are active for this pipeline
uint32_t active_shaders;
uint32_t duplicate_shaders;
// Capture which slots (set#->bindings) are actually used by the shaders of this pipeline
std::unordered_map<uint32_t, BindingReqMap> active_slots;
uint32_t max_active_slot; // the highest set number in active_slots for pipeline layout compatibility checks
// Additional metadata needed by pipeline_state initialization and validation
std::vector<StageState> stage_state;
// Vtx input info (if any)
std::vector<VkVertexInputBindingDescription> vertex_binding_descriptions_;
std::vector<VkVertexInputAttributeDescription> vertex_attribute_descriptions_;
std::vector<VkDeviceSize> vertex_attribute_alignments_;
std::unordered_map<uint32_t, uint32_t> vertex_binding_to_index_map_;
std::vector<VkPipelineColorBlendAttachmentState> attachments;
bool blendConstantsEnabled; // Blend constants enabled for any attachments
std::shared_ptr<const PIPELINE_LAYOUT_STATE> pipeline_layout;
VkPrimitiveTopology topology_at_rasterizer;
// Default constructor
PIPELINE_STATE()
: pipeline{},
graphicsPipelineCI{},
computePipelineCI{},
raytracingPipelineCI{},
rp_state(nullptr),
active_shaders(0),
duplicate_shaders(0),
active_slots(),
max_active_slot(0),
vertex_binding_descriptions_(),
vertex_attribute_descriptions_(),
vertex_binding_to_index_map_(),
attachments(),
blendConstantsEnabled(false),
pipeline_layout(),
topology_at_rasterizer{} {}
void reset() {
VkGraphicsPipelineCreateInfo emptyGraphicsCI = {};
graphicsPipelineCI.initialize(&emptyGraphicsCI, false, false);
VkComputePipelineCreateInfo emptyComputeCI = {};
computePipelineCI.initialize(&emptyComputeCI);
VkRayTracingPipelineCreateInfoNV emptyRayTracingCI = {};
raytracingPipelineCI.initialize(&emptyRayTracingCI);
stage_state.clear();
}
void initGraphicsPipeline(const ValidationStateTracker *state_data, const VkGraphicsPipelineCreateInfo *pCreateInfo,
std::shared_ptr<const RENDER_PASS_STATE> &&rpstate);
void initComputePipeline(const ValidationStateTracker *state_data, const VkComputePipelineCreateInfo *pCreateInfo);
void initRayTracingPipelineNV(const ValidationStateTracker *state_data, const VkRayTracingPipelineCreateInfoNV *pCreateInfo);
inline VkPipelineBindPoint getPipelineType() const {
if (graphicsPipelineCI.sType == VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO)
return VK_PIPELINE_BIND_POINT_GRAPHICS;
else if (computePipelineCI.sType == VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO)
return VK_PIPELINE_BIND_POINT_COMPUTE;
else if (raytracingPipelineCI.sType == VK_STRUCTURE_TYPE_RAY_TRACING_PIPELINE_CREATE_INFO_NV)
return VK_PIPELINE_BIND_POINT_RAY_TRACING_NV;
else
return VK_PIPELINE_BIND_POINT_MAX_ENUM;
}
inline VkPipelineCreateFlags getPipelineCreateFlags() const {
if (graphicsPipelineCI.sType == VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO)
return graphicsPipelineCI.flags;
else if (computePipelineCI.sType == VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO)
return computePipelineCI.flags;
else if (raytracingPipelineCI.sType == VK_STRUCTURE_TYPE_RAY_TRACING_PIPELINE_CREATE_INFO_NV)
return raytracingPipelineCI.flags;
else
return 0;
}
};
// Track last states that are bound per pipeline bind point (Gfx & Compute)
struct LAST_BOUND_STATE {
LAST_BOUND_STATE() { reset(); } // must define default constructor for portability reasons
PIPELINE_STATE *pipeline_state;
VkPipelineLayout pipeline_layout;
std::unique_ptr<cvdescriptorset::DescriptorSet> push_descriptor_set;
// Ordered bound set tracking where index is set# that given set is bound to
struct PER_SET {
PER_SET()
: bound_descriptor_set(nullptr),
compat_id_for_set(0),
validated_set(nullptr),
validated_set_change_count(~0ULL),
validated_set_image_layout_change_count(~0ULL),
validated_set_binding_req_map() {}
cvdescriptorset::DescriptorSet *bound_descriptor_set;
// one dynamic offset per dynamic descriptor bound to this CB
std::vector<uint32_t> dynamicOffsets;
PipelineLayoutCompatId compat_id_for_set;
// Cache most recently validated descriptor state for ValidateCmdBufDrawState/UpdateDrawState
const cvdescriptorset::DescriptorSet *validated_set;
uint64_t validated_set_change_count;
uint64_t validated_set_image_layout_change_count;
BindingReqMap validated_set_binding_req_map;
};
std::vector<PER_SET> per_set;
void reset() {
pipeline_state = nullptr;
pipeline_layout = VK_NULL_HANDLE;
push_descriptor_set = nullptr;
per_set.clear();
}
void UnbindAndResetPushDescriptorSet(cvdescriptorset::DescriptorSet *ds) {
if (push_descriptor_set) {
for (std::size_t i = 0; i < per_set.size(); i++) {
if (per_set[i].bound_descriptor_set == push_descriptor_set.get()) {
per_set[i].bound_descriptor_set = nullptr;
}
}
}
push_descriptor_set.reset(ds);
}
};
static inline bool CompatForSet(uint32_t set, const LAST_BOUND_STATE &a, const std::vector<PipelineLayoutCompatId> &b) {
bool result = (set < a.per_set.size()) && (set < b.size()) && (a.per_set[set].compat_id_for_set == b[set]);
return result;
}
static inline bool CompatForSet(uint32_t set, const PIPELINE_LAYOUT_STATE *a, const PIPELINE_LAYOUT_STATE *b) {
// Intentionally have a result variable to simplify debugging
bool result = a && b && (set < a->compat_for_set.size()) && (set < b->compat_for_set.size()) &&
(a->compat_for_set[set] == b->compat_for_set[set]);
return result;
}
// Types to store queue family ownership (QFO) Transfers
// Common to image and buffer memory barriers
template <typename Handle, typename Barrier>
struct QFOTransferBarrierBase {
using HandleType = Handle;
using BarrierType = Barrier;
struct Tag {};
HandleType handle = VK_NULL_HANDLE;
uint32_t srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
uint32_t dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
QFOTransferBarrierBase() = default;
QFOTransferBarrierBase(const BarrierType &barrier, const HandleType &resource_handle)
: handle(resource_handle),
srcQueueFamilyIndex(barrier.srcQueueFamilyIndex),
dstQueueFamilyIndex(barrier.dstQueueFamilyIndex) {}
hash_util::HashCombiner base_hash_combiner() const {
hash_util::HashCombiner hc;
hc << srcQueueFamilyIndex << dstQueueFamilyIndex << handle;
return hc;
}
bool operator==(const QFOTransferBarrierBase &rhs) const {
return (srcQueueFamilyIndex == rhs.srcQueueFamilyIndex) && (dstQueueFamilyIndex == rhs.dstQueueFamilyIndex) &&
(handle == rhs.handle);
}
};
template <typename Barrier>
struct QFOTransferBarrier {};
// Image barrier specific implementation
template <>
struct QFOTransferBarrier<VkImageMemoryBarrier> : public QFOTransferBarrierBase<VkImage, VkImageMemoryBarrier> {
using BaseType = QFOTransferBarrierBase<VkImage, VkImageMemoryBarrier>;
VkImageLayout oldLayout = VK_IMAGE_LAYOUT_UNDEFINED;
VkImageLayout newLayout = VK_IMAGE_LAYOUT_UNDEFINED;
VkImageSubresourceRange subresourceRange;
QFOTransferBarrier() = default;
QFOTransferBarrier(const BarrierType &barrier)
: BaseType(barrier, barrier.image),
oldLayout(barrier.oldLayout),
newLayout(barrier.newLayout),
subresourceRange(barrier.subresourceRange) {}
size_t hash() const {
// Ignoring the layout information for the purpose of the hash, as we're interested in QFO release/acquisition w.r.t.
// the subresource affected, an layout transitions are current validated on another path
auto hc = base_hash_combiner() << subresourceRange;
return hc.Value();
}
bool operator==(const QFOTransferBarrier<BarrierType> &rhs) const {
// Ignoring layout w.r.t. equality. See comment in hash above.
return (static_cast<BaseType>(*this) == static_cast<BaseType>(rhs)) && (subresourceRange == rhs.subresourceRange);
}
// TODO: codegen a comprehensive complie time type -> string (and or other traits) template family
static const char *BarrierName() { return "VkImageMemoryBarrier"; }
static const char *HandleName() { return "VkImage"; }
// UNASSIGNED-VkImageMemoryBarrier-image-00001 QFO transfer image barrier must not duplicate QFO recorded in command buffer
static const char *ErrMsgDuplicateQFOInCB() { return "UNASSIGNED-VkImageMemoryBarrier-image-00001"; }
// UNASSIGNED-VkImageMemoryBarrier-image-00002 QFO transfer image barrier must not duplicate QFO submitted in batch
static const char *ErrMsgDuplicateQFOInSubmit() { return "UNASSIGNED-VkImageMemoryBarrier-image-00002"; }
// UNASSIGNED-VkImageMemoryBarrier-image-00003 QFO transfer image barrier must not duplicate QFO submitted previously
static const char *ErrMsgDuplicateQFOSubmitted() { return "UNASSIGNED-VkImageMemoryBarrier-image-00003"; }
// UNASSIGNED-VkImageMemoryBarrier-image-00004 QFO acquire image barrier must have matching QFO release submitted previously
static const char *ErrMsgMissingQFOReleaseInSubmit() { return "UNASSIGNED-VkImageMemoryBarrier-image-00004"; }
};
// Buffer barrier specific implementation
template <>
struct QFOTransferBarrier<VkBufferMemoryBarrier> : public QFOTransferBarrierBase<VkBuffer, VkBufferMemoryBarrier> {
using BaseType = QFOTransferBarrierBase<VkBuffer, VkBufferMemoryBarrier>;
VkDeviceSize offset = 0;
VkDeviceSize size = 0;
QFOTransferBarrier(const VkBufferMemoryBarrier &barrier)
: BaseType(barrier, barrier.buffer), offset(barrier.offset), size(barrier.size) {}
size_t hash() const {
auto hc = base_hash_combiner() << offset << size;
return hc.Value();
}
bool operator==(const QFOTransferBarrier<BarrierType> &rhs) const {
return (static_cast<BaseType>(*this) == static_cast<BaseType>(rhs)) && (offset == rhs.offset) && (size == rhs.size);
}
static const char *BarrierName() { return "VkBufferMemoryBarrier"; }
static const char *HandleName() { return "VkBuffer"; }
// UNASSIGNED-VkImageMemoryBarrier-buffer-00001 QFO transfer buffer barrier must not duplicate QFO recorded in command buffer
static const char *ErrMsgDuplicateQFOInCB() { return "UNASSIGNED-VkBufferMemoryBarrier-buffer-00001"; }
// UNASSIGNED-VkBufferMemoryBarrier-buffer-00002 QFO transfer buffer barrier must not duplicate QFO submitted in batch
static const char *ErrMsgDuplicateQFOInSubmit() { return "UNASSIGNED-VkBufferMemoryBarrier-buffer-00002"; }
// UNASSIGNED-VkBufferMemoryBarrier-buffer-00003 QFO transfer buffer barrier must not duplicate QFO submitted previously
static const char *ErrMsgDuplicateQFOSubmitted() { return "UNASSIGNED-VkBufferMemoryBarrier-buffer-00003"; }
// UNASSIGNED-VkBufferMemoryBarrier-buffer-00004 QFO acquire buffer barrier must have matching QFO release submitted previously
static const char *ErrMsgMissingQFOReleaseInSubmit() { return "UNASSIGNED-VkBufferMemoryBarrier-buffer-00004"; }
};
template <typename Barrier>
using QFOTransferBarrierHash = hash_util::HasHashMember<QFOTransferBarrier<Barrier>>;
// Command buffers store the set of barriers recorded
template <typename Barrier>
using QFOTransferBarrierSet = std::unordered_set<QFOTransferBarrier<Barrier>, QFOTransferBarrierHash<Barrier>>;
template <typename Barrier>
struct QFOTransferBarrierSets {
QFOTransferBarrierSet<Barrier> release;
QFOTransferBarrierSet<Barrier> acquire;
void Reset() {
acquire.clear();
release.clear();
}
};
// The layer_data stores the map of pending release barriers
template <typename Barrier>
using GlobalQFOTransferBarrierMap =
std::unordered_map<typename QFOTransferBarrier<Barrier>::HandleType, QFOTransferBarrierSet<Barrier>>;
// Submit queue uses the Scoreboard to track all release/acquire operations in a batch.
template <typename Barrier>
using QFOTransferCBScoreboard =
std::unordered_map<QFOTransferBarrier<Barrier>, const CMD_BUFFER_STATE *, QFOTransferBarrierHash<Barrier>>;
template <typename Barrier>
struct QFOTransferCBScoreboards {
QFOTransferCBScoreboard<Barrier> acquire;
QFOTransferCBScoreboard<Barrier> release;
};
typedef std::map<QueryObject, QueryState> QueryMap;
typedef std::map<QueryObjectPass, QueryState> QueryPassMap;
typedef std::unordered_map<VkEvent, VkPipelineStageFlags> EventToStageMap;
// Cmd Buffer Wrapper Struct - TODO : This desperately needs its own class
struct CMD_BUFFER_STATE : public BASE_NODE {
VkCommandBuffer commandBuffer;
VkCommandBufferAllocateInfo createInfo = {};
VkCommandBufferBeginInfo beginInfo;
VkCommandBufferInheritanceInfo inheritanceInfo;
VkDevice device; // device this CB belongs to
std::shared_ptr<const COMMAND_POOL_STATE> command_pool;
bool hasDrawCmd;
bool hasTraceRaysCmd;
bool hasBuildAccelerationStructureCmd;
bool hasDispatchCmd;
CB_STATE state; // Track cmd buffer update state
uint64_t commandCount; // Number of commands recorded
uint64_t submitCount; // Number of times CB has been submitted
typedef uint64_t ImageLayoutUpdateCount;
ImageLayoutUpdateCount image_layout_change_count; // The sequence number for changes to image layout (for cached validation)
CBStatusFlags status; // Track status of various bindings on cmd buffer
CBStatusFlags static_status; // All state bits provided by current graphics pipeline
// rather than dynamic state
// Currently storing "lastBound" objects on per-CB basis
// long-term may want to create caches of "lastBound" states and could have
// each individual CMD_NODE referencing its own "lastBound" state
// Store last bound state for Gfx & Compute pipeline bind points
std::map<uint32_t, LAST_BOUND_STATE> lastBound;
using Bindings = std::map<uint32_t, descriptor_req>;
using Pipelines_Bindings = std::map<VkPipeline, Bindings>;
std::unordered_map<VkDescriptorSet, Pipelines_Bindings> validate_descriptorsets_in_queuesubmit;
uint32_t viewportMask;
uint32_t scissorMask;
uint32_t initial_device_mask;
VkRenderPassBeginInfo activeRenderPassBeginInfo;
RENDER_PASS_STATE *activeRenderPass;
VkSubpassContents activeSubpassContents;
uint32_t active_render_pass_device_mask;
uint32_t activeSubpass;
VkFramebuffer activeFramebuffer;
std::unordered_set<VkFramebuffer> framebuffers;
// Unified data structs to track objects bound to this command buffer as well as object
// dependencies that have been broken : either destroyed objects, or updated descriptor sets
std::vector<VulkanTypedHandle> object_bindings;
std::vector<VulkanTypedHandle> broken_bindings;
QFOTransferBarrierSets<VkBufferMemoryBarrier> qfo_transfer_buffer_barriers;
QFOTransferBarrierSets<VkImageMemoryBarrier> qfo_transfer_image_barriers;
std::unordered_set<VkEvent> waitedEvents;
std::vector<VkEvent> writeEventsBeforeWait;
std::vector<VkEvent> events;
std::unordered_set<QueryObject> activeQueries;
std::unordered_set<QueryObject> startedQueries;
typedef std::unordered_map<VkImage, std::unique_ptr<ImageSubresourceLayoutMap>> ImageLayoutMap;
ImageLayoutMap image_layout_map;
CBVertexBufferBindingInfo current_vertex_buffer_binding_info;
bool vertex_buffer_used; // Track for perf warning to make sure any bound vtx buffer used
VkCommandBuffer primaryCommandBuffer;
// If primary, the secondary command buffers we will call.
// If secondary, the primary command buffers we will be called by.
std::unordered_set<CMD_BUFFER_STATE *> linkedCommandBuffers;
// Validation functions run at primary CB queue submit time
std::vector<std::function<bool(const ValidationStateTracker *device_data, const class QUEUE_STATE *queue_state)>>
queue_submit_functions;
// Validation functions run when secondary CB is executed in primary
std::vector<std::function<bool(const CMD_BUFFER_STATE *, VkFramebuffer)>> cmd_execute_commands_functions;
std::vector<
std::function<bool(const ValidationStateTracker *device_data, bool do_validate, EventToStageMap *localEventToStageMap)>>
eventUpdates;
std::vector<std::function<bool(const ValidationStateTracker *device_data, bool do_validate, QueryMap *localQueryToStateMap)>>
queryUpdates;
std::unordered_set<cvdescriptorset::DescriptorSet *> validated_descriptor_sets;
// Contents valid only after an index buffer is bound (CBSTATUS_INDEX_BUFFER_BOUND set)
IndexBufferBinding index_buffer_binding;
bool performance_lock_acquired = false;
bool performance_lock_released = false;
// Cache of current insert label...
LoggingLabel debug_label;
std::vector<uint8_t> push_constant_data;
PushConstantRangesId push_constant_data_ranges;
};
static inline const QFOTransferBarrierSets<VkImageMemoryBarrier> &GetQFOBarrierSets(
const CMD_BUFFER_STATE *cb, const QFOTransferBarrier<VkImageMemoryBarrier>::Tag &type_tag) {
return cb->qfo_transfer_image_barriers;
}
static inline const QFOTransferBarrierSets<VkBufferMemoryBarrier> &GetQFOBarrierSets(
const CMD_BUFFER_STATE *cb, const QFOTransferBarrier<VkBufferMemoryBarrier>::Tag &type_tag) {
return cb->qfo_transfer_buffer_barriers;
}
static inline QFOTransferBarrierSets<VkImageMemoryBarrier> &GetQFOBarrierSets(
CMD_BUFFER_STATE *cb, const QFOTransferBarrier<VkImageMemoryBarrier>::Tag &type_tag) {
return cb->qfo_transfer_image_barriers;
}
static inline QFOTransferBarrierSets<VkBufferMemoryBarrier> &GetQFOBarrierSets(
CMD_BUFFER_STATE *cb, const QFOTransferBarrier<VkBufferMemoryBarrier>::Tag &type_tag) {
return cb->qfo_transfer_buffer_barriers;
}
struct SEMAPHORE_WAIT {
VkSemaphore semaphore;
VkQueue queue;
uint64_t seq;
};
struct CB_SUBMISSION {
CB_SUBMISSION(std::vector<VkCommandBuffer> const &cbs, std::vector<SEMAPHORE_WAIT> const &waitSemaphores,
std::vector<VkSemaphore> const &signalSemaphores, std::vector<VkSemaphore> const &externalSemaphores,
VkFence fence, uint32_t perf_submit_pass)
: cbs(cbs),
waitSemaphores(waitSemaphores),
signalSemaphores(signalSemaphores),
externalSemaphores(externalSemaphores),
fence(fence),
perf_submit_pass(perf_submit_pass) {}
std::vector<VkCommandBuffer> cbs;
std::vector<SEMAPHORE_WAIT> waitSemaphores;
std::vector<VkSemaphore> signalSemaphores;
std::vector<VkSemaphore> externalSemaphores;
VkFence fence;
uint32_t perf_submit_pass;
};
struct IMAGE_LAYOUT_STATE {
VkImageLayout layout;
VkFormat format;
};
struct MT_FB_ATTACHMENT_INFO {
IMAGE_VIEW_STATE *view_state;
VkImage image;
};
class FRAMEBUFFER_STATE : public BASE_NODE {
public:
VkFramebuffer framebuffer;
safe_VkFramebufferCreateInfo createInfo;
std::shared_ptr<const RENDER_PASS_STATE> rp_state;
FRAMEBUFFER_STATE(VkFramebuffer fb, const VkFramebufferCreateInfo *pCreateInfo, std::shared_ptr<RENDER_PASS_STATE> &&rpstate)
: framebuffer(fb), createInfo(pCreateInfo), rp_state(rpstate){};
};
struct SHADER_MODULE_STATE;
struct DeviceExtensions;
struct DeviceFeatures {
VkPhysicalDeviceFeatures core;
VkPhysicalDeviceDescriptorIndexingFeaturesEXT descriptor_indexing;
VkPhysicalDevice8BitStorageFeaturesKHR eight_bit_storage;
VkPhysicalDeviceExclusiveScissorFeaturesNV exclusive_scissor;
VkPhysicalDeviceShadingRateImageFeaturesNV shading_rate_image;
VkPhysicalDeviceMeshShaderFeaturesNV mesh_shader;
VkPhysicalDeviceInlineUniformBlockFeaturesEXT inline_uniform_block;
VkPhysicalDeviceTransformFeedbackFeaturesEXT transform_feedback_features;
VkPhysicalDeviceFloat16Int8FeaturesKHR float16_int8;
VkPhysicalDeviceVertexAttributeDivisorFeaturesEXT vtx_attrib_divisor_features;
VkPhysicalDeviceUniformBufferStandardLayoutFeaturesKHR uniform_buffer_standard_layout;
VkPhysicalDeviceScalarBlockLayoutFeaturesEXT scalar_block_layout_features;
VkPhysicalDeviceBufferDeviceAddressFeaturesKHR buffer_device_address;
VkPhysicalDeviceBufferDeviceAddressFeaturesEXT buffer_device_address_ext;
VkPhysicalDeviceCooperativeMatrixFeaturesNV cooperative_matrix_features;
VkPhysicalDeviceHostQueryResetFeaturesEXT host_query_reset_features;
VkPhysicalDeviceComputeShaderDerivativesFeaturesNV compute_shader_derivatives_features;
VkPhysicalDeviceFragmentShaderBarycentricFeaturesNV fragment_shader_barycentric_features;
VkPhysicalDeviceShaderImageFootprintFeaturesNV shader_image_footprint_features;
VkPhysicalDeviceFragmentShaderInterlockFeaturesEXT fragment_shader_interlock_features;
VkPhysicalDeviceShaderDemoteToHelperInvocationFeaturesEXT demote_to_helper_invocation_features;
VkPhysicalDeviceTexelBufferAlignmentFeaturesEXT texel_buffer_alignment_features;
VkPhysicalDeviceImagelessFramebufferFeaturesKHR imageless_framebuffer_features;
VkPhysicalDevicePipelineExecutablePropertiesFeaturesKHR pipeline_exe_props_features;
VkPhysicalDeviceDedicatedAllocationImageAliasingFeaturesNV dedicated_allocation_image_aliasing_features;
VkPhysicalDeviceShaderSubgroupExtendedTypesFeaturesKHR subgroup_extended_types_features;
VkPhysicalDeviceSeparateDepthStencilLayoutsFeaturesKHR separate_depth_stencil_layouts_features;
VkPhysicalDevicePerformanceQueryFeaturesKHR performance_query_features;
VkPhysicalDeviceTimelineSemaphoreFeaturesKHR timeline_semaphore_features;
VkPhysicalDeviceCoherentMemoryFeaturesAMD device_coherent_memory_features;
};
enum RenderPassCreateVersion { RENDER_PASS_VERSION_1 = 0, RENDER_PASS_VERSION_2 = 1 };
enum CommandVersion { CMD_VERSION_1 = 0, CMD_VERSION_2 = 1 };
enum BarrierOperationsType {
kAllAcquire, // All Barrier operations are "ownership acquire" operations
kAllRelease, // All Barrier operations are "ownership release" operations
kGeneral, // Either no ownership operations or a mix of ownership operation types and/or non-ownership operations
};
ImageSubresourceLayoutMap *GetImageSubresourceLayoutMap(CMD_BUFFER_STATE *cb_state, const IMAGE_STATE &image_state);
const ImageSubresourceLayoutMap *GetImageSubresourceLayoutMap(const CMD_BUFFER_STATE *cb_state, VkImage image);
#endif // CORE_VALIDATION_TYPES_H_