ganesh/vk/GrVkCaps.h

/*
 * Copyright 2015 Google Inc.
 *
 * Use of this source code is governed by a BSD-style license that can be
 * found in the LICENSE file.
 */

#ifndef GrVkCaps_DEFINED
#define GrVkCaps_DEFINED

#include "include/gpu/vk/GrVkTypes.h"
#include "include/private/base/SkTArray.h"
#include "include/private/base/SkTDArray.h"
#include "src/gpu/ganesh/GrCaps.h"

class GrVkRenderTarget;
enum class SkTextureCompressionType;

namespace skgpu {
class VulkanExtensions;
struct VulkanInterface;
}

/**
 * Stores some capabilities of a Vk backend.
 */
class GrVkCaps : public GrCaps {
public:
    /**
     * Creates a GrVkCaps that is set such that nothing is supported. The init function should
     * be called to fill out the caps.
     */
    GrVkCaps(const GrContextOptions&,
             const skgpu::VulkanInterface*,
             VkPhysicalDevice,
             const VkPhysicalDeviceFeatures2&,
             uint32_t instanceVersion,
             uint32_t physicalDeviceVersion,
             const skgpu::VulkanExtensions&,
             skgpu::Protected);

    bool isFormatSRGB(const GrBackendFormat&) const override;

    bool isFormatTexturable(const GrBackendFormat&, GrTextureType) const override;
    bool isVkFormatTexturable(VkFormat) const;

    bool isFormatCopyable(const GrBackendFormat&) const override { return true; }

    bool isFormatAsColorTypeRenderable(GrColorType ct,
                                       const GrBackendFormat& format,
                                       int sampleCount = 1) const override;
    bool isFormatRenderable(const GrBackendFormat& format, int sampleCount) const override;
    bool isFormatRenderable(VkFormat, int sampleCount) const;

    int getRenderTargetSampleCount(int requestedCount, const GrBackendFormat&) const override;
    int getRenderTargetSampleCount(int requestedCount, VkFormat) const;

    int maxRenderTargetSampleCount(const GrBackendFormat&) const override;
    int maxRenderTargetSampleCount(VkFormat format) const;

    SupportedWrite supportedWritePixelsColorType(GrColorType surfaceColorType,
                                                 const GrBackendFormat& surfaceFormat,
                                                 GrColorType srcColorType) const override;

    SurfaceReadPixelsSupport surfaceSupportsReadPixels(const GrSurface*) const override;

    bool isVkFormatTexturableLinearly(VkFormat format) const {
        return SkToBool(FormatInfo::kTexturable_Flag & this->getFormatInfo(format).fLinearFlags);
    }

    bool formatCanBeDstofBlit(VkFormat format, bool linearTiled) const {
        const FormatInfo& info = this->getFormatInfo(format);
        const uint16_t& flags = linearTiled ? info.fLinearFlags : info.fOptimalFlags;
        return SkToBool(FormatInfo::kBlitDst_Flag & flags);
    }

    bool formatCanBeSrcofBlit(VkFormat format, bool linearTiled) const {
        const FormatInfo& info = this->getFormatInfo(format);
        const uint16_t& flags = linearTiled ? info.fLinearFlags : info.fOptimalFlags;
        return SkToBool(FormatInfo::kBlitSrc_Flag & flags);
    }

    // Gets the GrColorType that should be used to transfer data in/out of a transfer buffer to
    // write/read data when using a VkFormat with a specified color type.
    GrColorType transferColorType(VkFormat, GrColorType surfaceColorType) const;

    // On some GPUs (Windows Nvidia and Imagination) calls to QueueWaitIdle return before actually
    // signalling the fences on the command buffers even though they have completed. This causes
    // issues when then deleting the command buffers. Therefore we additionally will call
    // vkWaitForFences on each outstanding command buffer to make sure the driver signals the fence.
    bool mustSyncCommandBuffersWithQueue() const { return fMustSyncCommandBuffersWithQueue; }

    // Returns true if we should always make dedicated allocations for VkImages.
    bool shouldAlwaysUseDedicatedImageMemory() const {
        return fShouldAlwaysUseDedicatedImageMemory;
    }

    // Always use a transfer buffer instead of vkCmdUpdateBuffer to upload data to a VkBuffer.
    bool avoidUpdateBuffers() const { return fAvoidUpdateBuffers; }

    /**
     * Returns both a supported and most preferred stencil format to use in draws.
     */
    VkFormat preferredStencilFormat() const { return fPreferredStencilFormat; }

    // Returns total number of bits used by stencil + depth + padding
    static int GetStencilFormatTotalBitCount(VkFormat format) {
        switch (format) {
            case VK_FORMAT_S8_UINT:
                return 8;
            case VK_FORMAT_D24_UNORM_S8_UINT:
                return 32;
            case VK_FORMAT_D32_SFLOAT_S8_UINT:
                // can optionally have 24 unused bits at the end so we assume the total bits is 64.
                return 64;
            default:
                SkASSERT(false);
                return 0;
        }
    }

    // Returns whether the device supports VK_KHR_Swapchain. Internally Skia never uses any of the
    // swapchain functions, but we may need to transition to and from the
    // VK_IMAGE_LAYOUT_PRESENT_SRC_KHR image layout, so we must know whether that layout is
    // supported.
    bool supportsSwapchain() const { return fSupportsSwapchain; }

    // Returns whether the device supports the ability to extend VkPhysicalDeviceProperties struct.
    bool supportsPhysicalDeviceProperties2() const { return fSupportsPhysicalDeviceProperties2; }
    // Returns whether the device supports the ability to extend VkMemoryRequirements struct.
    bool supportsMemoryRequirements2() const { return fSupportsMemoryRequirements2; }

    // Returns whether the device supports the ability to extend the vkBindMemory call.
    bool supportsBindMemory2() const { return fSupportsBindMemory2; }

    // Returns whether or not the device suports the various API maintenance fixes to Vulkan 1.0. In
    // Vulkan 1.1 all these maintenance are part of the core spec.
    bool supportsMaintenance1() const { return fSupportsMaintenance1; }
    bool supportsMaintenance2() const { return fSupportsMaintenance2; }
    bool supportsMaintenance3() const { return fSupportsMaintenance3; }

    // Returns true if the device supports passing in a flag to say we are using dedicated GPU when
    // allocating memory. For some devices this allows them to return more optimized memory knowning
    // they will never need to suballocate amonst multiple objects.
    bool supportsDedicatedAllocation() const { return fSupportsDedicatedAllocation; }

    // Returns true if the device supports importing of external memory into Vulkan memory.
    bool supportsExternalMemory() const { return fSupportsExternalMemory; }
    // Returns true if the device supports importing Android hardware buffers into Vulkan memory.
    bool supportsAndroidHWBExternalMemory() const { return fSupportsAndroidHWBExternalMemory; }

    // Returns true if it supports ycbcr conversion for samplers
    bool supportsYcbcrConversion() const { return fSupportsYcbcrConversion; }

    // Returns the number of descriptor slots used by immutable ycbcr VkImages.
    //
    // TODO: We should update this to return a count for a specific format or external format. We
    // can use vkGetPhysicalDeviceImageFormatProperties2 with a
    // VkSamplerYcbcrConversionImageFormatProperties to query this. However, right now that call
    // does not support external android formats which is where the majority of ycbcr images are
    // coming from. So for now we stay safe and always return 3 here which is the max value that the
    // count could be for any format.
    uint32_t ycbcrCombinedImageSamplerDescriptorCount() const {
        return 3;
    }

    // Returns true if the VK_EXT_image_drm_format_modifier is enabled.
    bool supportsDRMFormatModifiers() const { return fSupportsDRMFormatModifiers; }

    bool supportsDeviceFaultInfo() const { return fSupportsDeviceFaultInfo; }

    // Returns whether we prefer to record draws directly into a primary command buffer.
    bool preferPrimaryOverSecondaryCommandBuffers() const {
        return fPreferPrimaryOverSecondaryCommandBuffers;
    }

    int maxPerPoolCachedSecondaryCommandBuffers() const {
        return fMaxPerPoolCachedSecondaryCommandBuffers;
    }

    uint32_t maxInputAttachmentDescriptors() const { return fMaxInputAttachmentDescriptors; }

    float maxSamplerAnisotropy() const { return fMaxSamplerAnisotropy; }

    bool mustInvalidatePrimaryCmdBufferStateAfterClearAttachments() const {
        return fMustInvalidatePrimaryCmdBufferStateAfterClearAttachments;
    }

    // Returns whether a pure GPU accessible buffer is more performant to read than a buffer that is
    // also host visible. If so then in some cases we may prefer the cost of doing a copy to the
    // buffer. This typically would only be the case for buffers that are written once and read
    // many times on the gpu.
    bool gpuOnlyBuffersMorePerformant() const { return fGpuOnlyBuffersMorePerformant; }

    // For our CPU write and GPU read buffers (vertex, uniform, etc.), should we keep these buffers
    // persistently mapped. In general the answer will be yes. The main case we don't do this is
    // when using special memory that is DEVICE_LOCAL and HOST_VISIBLE on discrete GPUs.
    bool shouldPersistentlyMapCpuToGpuBuffers() const {
        return fShouldPersistentlyMapCpuToGpuBuffers;
    }

    // The max draw count that can be passed into indirect draw calls.
    uint32_t  maxDrawIndirectDrawCount() const { return fMaxDrawIndirectDrawCount; }

    /**
     * Helpers used by canCopySurface. In all cases if the SampleCnt parameter is zero that means
     * the surface is not a render target, otherwise it is the number of samples in the render
     * target.
     */
    bool canCopyImage(VkFormat dstFormat,
                      int dstSampleCnt,
                      bool dstHasYcbcr,
                      VkFormat srcFormat,
                      int srcSamplecnt,
                      bool srcHasYcbcr) const;

    bool canCopyAsBlit(VkFormat dstConfig,
                       int dstSampleCnt,
                       bool dstIsLinear,
                       bool dstHasYcbcr,
                       VkFormat srcConfig,
                       int srcSampleCnt,
                       bool srcIsLinear,
                       bool srcHasYcbcr) const;

    bool canCopyAsResolve(VkFormat dstConfig,
                          int dstSampleCnt,
                          bool dstHasYcbcr,
                          VkFormat srcConfig,
                          int srcSamplecnt,
                          bool srcHasYcbcr) const;

    GrBackendFormat getBackendFormatFromCompressionType(SkTextureCompressionType) const override;

    VkFormat getFormatFromColorType(GrColorType colorType) const {
        int idx = static_cast<int>(colorType);
        return fColorTypeToFormatTable[idx];
    }

    skgpu::Swizzle getWriteSwizzle(const GrBackendFormat&, GrColorType) const override;

    uint64_t computeFormatKey(const GrBackendFormat&) const override;

    int getFragmentUniformBinding() const;
    int getFragmentUniformSet() const;

    void addExtraSamplerKey(skgpu::KeyBuilder*,
                            GrSamplerState,
                            const GrBackendFormat&) const override;

    GrProgramDesc makeDesc(GrRenderTarget*,
                           const GrProgramInfo&,
                           ProgramDescOverrideFlags) const override;

    GrInternalSurfaceFlags getExtraSurfaceFlagsForDeferredRT() const override;

    VkShaderStageFlags getPushConstantStageFlags() const;

    bool mustLoadFullImageWithDiscardableMSAA() const {
        return fMustLoadFullImageWithDiscardableMSAA;
    }
    bool supportsDiscardableMSAAForDMSAA() const { return fSupportsDiscardableMSAAForDMSAA; }
    bool renderTargetSupportsDiscardableMSAA(const GrVkRenderTarget*) const;
    bool programInfoWillUseDiscardableMSAA(const GrProgramInfo&) const;

    bool dmsaaResolveCanBeUsedAsTextureInSameRenderPass() const override { return false; }

    bool supportsMemorylessAttachments() const { return fSupportsMemorylessAttachments; }

#if defined(GR_TEST_UTILS)
    std::vector<GrTest::TestFormatColorTypeCombination> getTestingCombinations() const override;
#endif

private:
    enum VkVendor {
        kAMD_VkVendor = 4098,
        kARM_VkVendor = 5045,
        kGoogle_VkVendor = 0x1AE0,
        kImagination_VkVendor = 4112,
        kIntel_VkVendor = 32902,
        kNvidia_VkVendor = 4318,
        kQualcomm_VkVendor = 20803,
    };

    enum class IntelGPUType {
        // 9th gen
        kSkyLake,

        // 11th gen
        kIceLake,

        // 12th gen
        kRocketLake,
        kTigerLake,
        kAlderLake,

        kOther
    };

    enum DeviceID {
        kSwiftshader_DeviceID = 0xC0DE, // As listed in Swiftshader code this may be a placeholder
                                        // value but works for now.
    };

    static IntelGPUType GetIntelGPUType(uint32_t deviceID);
    static int GetIntelGen(IntelGPUType type) {
        switch (type) {
            case IntelGPUType::kSkyLake:
                return 9;
            case IntelGPUType::kIceLake:
                return 11;
            case IntelGPUType::kRocketLake: // fall through
            case IntelGPUType::kTigerLake:  // fall through
            case IntelGPUType::kAlderLake:
                return 12;
            case IntelGPUType::kOther:
                // For now all our workaround checks are in the form of "if gen > some_value". So
                // we can return 0 for kOther which means we won't put in the new workaround for
                // older gens which is fine. If we stay on top of adding support for new gen
                // intel devices we shouldn't hit cases where we'd need to change this pattern.
                return 0;
        }
        SkUNREACHABLE;
    }

    void init(const GrContextOptions&,
              const skgpu::VulkanInterface*,
              VkPhysicalDevice,
              const VkPhysicalDeviceFeatures2&,
              uint32_t physicalDeviceVersion,
              const skgpu::VulkanExtensions&,
              GrProtected);
    void initGrCaps(const skgpu::VulkanInterface* vkInterface,
                    VkPhysicalDevice physDev,
                    const VkPhysicalDeviceProperties&,
                    const VkPhysicalDeviceMemoryProperties&,
                    const VkPhysicalDeviceFeatures2&,
                    const skgpu::VulkanExtensions&);
    void initShaderCaps(const VkPhysicalDeviceProperties&, const VkPhysicalDeviceFeatures2&);

    void initFormatTable(const GrContextOptions&,
                         const skgpu::VulkanInterface*,
                         VkPhysicalDevice,
                         const VkPhysicalDeviceProperties&,
                         const VkPhysicalDeviceFeatures2&,
                         const skgpu::VulkanExtensions&);
    void initStencilFormat(const skgpu::VulkanInterface* iface, VkPhysicalDevice physDev);

    void applyDriverCorrectnessWorkarounds(const VkPhysicalDeviceProperties&);

    bool onSurfaceSupportsWritePixels(const GrSurface*) const override;
    bool onCanCopySurface(const GrSurfaceProxy* dst, const SkIRect& dstRect,
                          const GrSurfaceProxy* src, const SkIRect& srcRect) const override;
    GrBackendFormat onGetDefaultBackendFormat(GrColorType) const override;

    bool onAreColorTypeAndFormatCompatible(GrColorType, const GrBackendFormat&) const override;

    SupportedRead onSupportedReadPixelsColorType(GrColorType, const GrBackendFormat&,
                                                 GrColorType) const override;

    skgpu::Swizzle onGetReadSwizzle(const GrBackendFormat&, GrColorType) const override;

    GrDstSampleFlags onGetDstSampleFlagsForProxy(const GrRenderTargetProxy*) const override;

    bool onSupportsDynamicMSAA(const GrRenderTargetProxy*) const override;

    // ColorTypeInfo for a specific format
    struct ColorTypeInfo {
        GrColorType fColorType = GrColorType::kUnknown;
        GrColorType fTransferColorType = GrColorType::kUnknown;
        enum {
            kUploadData_Flag = 0x1,
            // Does Ganesh itself support rendering to this colorType & format pair. Renderability
            // still additionally depends on if the format itself is renderable.
            kRenderable_Flag = 0x2,
            // Indicates that this colorType is supported only if we are wrapping a texture with
            // the given format and colorType. We do not allow creation with this pair.
            kWrappedOnly_Flag = 0x4,
        };
        uint32_t fFlags = 0;

        skgpu::Swizzle fReadSwizzle;
        skgpu::Swizzle fWriteSwizzle;
    };

    struct FormatInfo {
        uint32_t colorTypeFlags(GrColorType colorType) const {
            for (int i = 0; i < fColorTypeInfoCount; ++i) {
                if (fColorTypeInfos[i].fColorType == colorType) {
                    return fColorTypeInfos[i].fFlags;
                }
            }
            return 0;
        }

        void init(const GrContextOptions&,
                  const skgpu::VulkanInterface*,
                  VkPhysicalDevice,
                  const VkPhysicalDeviceProperties&,
                  VkFormat);
        static void InitFormatFlags(VkFormatFeatureFlags, uint16_t* flags);
        void initSampleCounts(const GrContextOptions&,
                              const skgpu::VulkanInterface*,
                              VkPhysicalDevice,
                              const VkPhysicalDeviceProperties&,
                              VkFormat);

        enum {
            kTexturable_Flag = 0x1,
            kRenderable_Flag = 0x2,
            kBlitSrc_Flag    = 0x4,
            kBlitDst_Flag    = 0x8,
        };

        uint16_t fOptimalFlags = 0;
        uint16_t fLinearFlags = 0;

        SkTDArray<int> fColorSampleCounts;

        std::unique_ptr<ColorTypeInfo[]> fColorTypeInfos;
        int fColorTypeInfoCount = 0;
    };
    static const size_t kNumVkFormats = 25;
    FormatInfo fFormatTable[kNumVkFormats];

    FormatInfo& getFormatInfo(VkFormat);
    const FormatInfo& getFormatInfo(VkFormat) const;

    VkFormat fColorTypeToFormatTable[kGrColorTypeCnt];
    void setColorType(GrColorType, std::initializer_list<VkFormat> formats);

    VkFormat fPreferredStencilFormat;

    skia_private::STArray<1, GrVkYcbcrConversionInfo> fYcbcrInfos;

    bool fMustSyncCommandBuffersWithQueue = false;
    bool fShouldAlwaysUseDedicatedImageMemory = false;

    bool fAvoidUpdateBuffers = false;

    bool fSupportsSwapchain = false;

    bool fSupportsPhysicalDeviceProperties2 = false;
    bool fSupportsMemoryRequirements2 = false;
    bool fSupportsBindMemory2 = false;
    bool fSupportsMaintenance1 = false;
    bool fSupportsMaintenance2 = false;
    bool fSupportsMaintenance3 = false;

    bool fSupportsDedicatedAllocation = false;
    bool fSupportsExternalMemory = false;
    bool fSupportsAndroidHWBExternalMemory = false;

    bool fSupportsYcbcrConversion = false;

    bool fSupportsDRMFormatModifiers = false;

    bool fSupportsDeviceFaultInfo = false;

    bool fPreferPrimaryOverSecondaryCommandBuffers = true;
    bool fMustInvalidatePrimaryCmdBufferStateAfterClearAttachments = false;

    bool fGpuOnlyBuffersMorePerformant = false;
    bool fShouldPersistentlyMapCpuToGpuBuffers = true;

    // We default this to 100 since we already cap the max render tasks at 100 before doing a
    // submission in the GrDrawingManager, so we shouldn't be going over 100 secondary command
    // buffers per primary anyways.
    int fMaxPerPoolCachedSecondaryCommandBuffers = 100;

    uint32_t fMaxInputAttachmentDescriptors = 0;

    float fMaxSamplerAnisotropy = 1.f;

    bool fMustLoadFullImageWithDiscardableMSAA = false;
    bool fSupportsDiscardableMSAAForDMSAA = true;
    bool fSupportsMemorylessAttachments = false;

    uint32_t fMaxDrawIndirectDrawCount = 0;

    using INHERITED = GrCaps;
};

#endif